You are on page 1of 96

GENERALIZED BASES FOR THE REAL NUiBEiS

J . A . F r i d y , Rutgers The State U n i v e r s i t y , N e w Brunswick, N e w Jersey

Throughout this paper { r i } ^ will denote a non-Increasing real number


sequence with limit zero; each of {k|}^ and {mj}^ 0 denotes a non-negative
integer sequence

S = V k.r.-

and

S* = V

m.r.

(finite or infinite). We shall consider the possibility of expressing each number x In the interval (-S*,S) in the form
CO

x =7^

a.r.

where each a. is an Integer satisfying -m. < a. < k..


In the classical n~scale number representation, each x in [ 0 , 1 ] can
be expressed in the above form, where n > 1, and r. = n" , k. = n - 1,
and m. = 0 for each ia

Previous generalizations ([6] and [8]) have con-

sidered only the expansion of positive numbers with certain restrictions on the
coefficient bounds {kj-f.
In this note we shall extend the previous work to include negative number
representations as well as relaxing the restrictions on the coefficients {a^}J.
We shall also consider the question of uniqueness of such representations and
the expansion of real numbers using a base sequence {rj}^ of both positive
and negative terms,
DEFINITION.

The sequence {r{}-f is a {k 9 m}-base for the Interval

(-S*,S) if for each x in (-*S*,S) there is an integeresequence


that
OO

(1)

x =

y^a.r. ,
LMJ

and

-m. ^ a. < k.

.11

193

for each I 0

{aj}^ such

194

GENERALIZ ED BASES

[ Oct

Our main purpose is to develop an explicit characterization of a {k,m}~


base; to this end we first consider the case where m. = 0 for each i; i e c ,
a {k,0}-base.
LEMMA0 The sequence {ri}^ is a {k, 0}-base for the interval (0,S)
if and only if

(2)

^ YJ k.r.

for each n0

n+i

Proof0

If (2) does not hold and

rn > x >

ZVl
n+i

for some n5 it is easily seen that x cannot be expressed in the form (1)0
Assume that (2) holds and let x be in (0,S), the conclusion being trivial
for x = 0. Let i(l) be the least positive integer such that r- n * =s x, and
choose a.yiv to be the &greatest integer
such that a.... < k._ x and a._ x r. A - v
s
i(l)

i(l)

i(l)

i(l) i(l)

^ x0

If a . / ^ r . , - , < x. we continue inductively:


J
1(1) 1(1)

'

Let i(n) be the least positive integer such that


n-i
3

<>

i(n) ^

~ E a i(p) r i(p)

and

i(n).>l(n-D

Choose a., v to be the greatest


integer
such that a.. , ^ k . .
&
&
i(n)

i(n)

i(n)v

and

n-i

<4>

i(n)ri(n) * X " E ai(p)ri(p)

In case equality does not hold in (4) for any n, we assert that

1966]

FOR THE R E A L NUMBERS

j^j

195

i(p> i(p>

Suppose, to the c o n t r a r y , that for s o m e positive


n

a., . r . , v ^ x - , for each n


x(p) i(p)

p=i
If r.> v < e it follows that
n-i
(6)

(a., x + l ) r . , x ^ x - y ^ a . , , r . , >
;
i(n)
i(n)
L^ i(p) i(p)
p=i

v ;

By the choice of a., , t h i s i m p l i e s that a

= k., ,;

furthermore,

yields

i(n)+i - r i(n) ~

" 2 J ai(p)ri(p) >


p=i

s o that i(n + 1) = i(n) + 1.' H e n c e ,

<7> E ^(p) r i(p) = E WKP) + E


p^i

p=i

VP " x

p=i(n)

Applying (2) to (7) we s e e that

n-i
(8)

i(n)-i *

" E ai(p)ri(p)
P=i

'

(6) a l s o

196

GENERALIZED BASES

[Oct

By the choice of i(n), (8) implies that i(n) - 1 = i(n - 1),

so that (8) can be

written as
n-2
(a.,
.,. + l ) r . , 1X < x - \^ a., ,r., .
v
i(n-l)
' i(n-l)
Z ^ i(p) i(p)
p=i
whence a., _- = k.

,
'

. Thus it is readily seen that for every n, i(n) =n

and a., , = k , which contradicts x < S; this establishes (5) and completes
the proof0
REMARK.

From this Lemma the following is clear:

If {r|}^ is a {k,0}-base for

[0,S) and N is apositive integer, then {r^} 0 ^

is a {k, 0}!!-base for the interval

L
Theorem 1

The sequence {ri}f

is a {k,m}-base for (-S*,S) if and

only if

(9)

<

\^

(k. 4- m.)r.

for each n

i=n+i
Proof

If (9) does not hold and

n >

x >

<ki +

m )r

i i

'

n+i
it follows easily from the Lemma that x - S* is in (-S*,S) but x - S* cannot
be expressed as in (1)
To show the sufficiency of (9) we first consider the case where S* is finite.

Let x be in (-S*,S).

By the Lemma, (9) guarantees a sequence {ai}J

1966 ]

FOE THE REAL NUMBERS

197

such that
oo

x + S* =

/LJ a i r i

and

0 < a. < k. + m.

for each i .

Letting b. = a. - m. , we have
OO

x = z^b.r. ,

and

-m. < b. < k.

for each i .

The case in which S is finite is proved similarly* If both S* and S


are infinite it follows immediately from the Lemma that every non-negative x
can be expressed as
oo

where 0 < a. < k., and every negative x can be so expressed with -m. ^ a.
< 0.
We now wish to establish conditions under which the representations in the
form (1) are unique. Since the common decimal expansion is not unique, and
this is the special case where r. = 10 , m. = 0, and k. = 9,
hope for total uniqueness in any non-trivial case

we cannot

Therefore we adopt a con-

vention similar to that used in identifying the decimal 0999 with .1000 ,
v i z . , we disallow a representation in which a. = k. for every i greater than
some nB Note that in the proof of the Lemma such representations were not
necessary 0

(This is also the reason that we did not consider the closed inter-

val ['0;9S] even when S was finite0


Theorem 2, The sequence {vj}t

yields exactly one {k,m}-base r e p r e -

sentation of each x in (-S*, S) if and only if

(10)

rn

2^
i=n+i

(\

+ m

i)' r i

for eacn

[Oct,

GENERALIZED BASES

198
Proof0

The sufficiency of (10) is fairly straightforward. Conversely, it

is easily seen that for unique representation it is necessary that gf (and S)


be finite,, Suppose that S* is finite and { r ^

satisfies (9) but not (10), Then

there exists an integer n and a number x such that

Y^ v(k.i + m.)r.
r n < x < 4-J
r i
n

n+i

Using the construction in the proof of the Lemma, we get a sequence {a^} 1
satisfying

5>i r i

and 0 ^ a. ^ k. + m.;
moreover,9 since r
9
i

< x. at least one of a^, 00 - 5 a


'

**

is non-zero. Taking b. = a. - m., we have

(11)

x - g* = ^ b . r . ,

where -m. < b. < k.

and for some i ^ n, b. 4= -m.,


On the other hand SrA
, is a {k
+ m9 0}J n +, i -base for the interval
L u +i
n

OO

>E( k i

+ m

i> r i)

n+i

by the Remark following the Lemma, This yields a second {k, m}-base representation: x - s =zZi d.r. , where d. = -m. for all i < n.
COROLLARY, The sequence {r^}^ yields a unique {k,m}-base representation of each x in (-S *, S) if and only if

1966 ]

FOE THE REAL NUMBERS


r n = (S + S * ) / n . * (1 + k. +

m.)

199
for each n .

Proof., This is straightforward induction using Theorem 2a


The foregoing theory can be used to consider representations of r e a l n u m bers in which the base sequence t r iJ 1 takes on both positive and negative
values. Let A and B be disjoint sets whose union is the set of positive integers, and let C and C R denote their respective characteristic functions,,
We shall use
loo

H)

C R (i)

r.
1

j1

as the base sequence


Theorem 3Q If {q,ifi is a positive integer sequence, then

11

iB

ieA

is a {q ? 0}-base for the interval

if and only if
00

(12)

" S

q r

ii

for each

n+i
Prooi
2.

A q.r.

Let k. = C A (i)q. and m. = C B (i)q t ,

= Ss

so that k. + m. = q.,

and 2. gQ-^ = Sf Thus by Theorem 1, (12) is equivalent

to {ri}^ being a {k, m}-base for (-S*SS)0 If (12) holds and x is in (-S*,S),
then
00

^ X

b r

ii ,

where

~CB(i)qi ~

i -

A(i)qi

200

GENERALIZED BASES

[Oct.

c B (i)
Taking a. = (-1)

(13)

b., we have

V UB(i)

and

0 < a. i <

The converse is proved similarly,,


REMARK*' It is clear that the representations in (13) are unique if and
only if equality holds in (12) for each n
A related problem is that of expressing a given number x in the form
OO

(14)

x = V^ e.r. ,

where

e. = 1 or -1

The following solution is proved using Theorem 1.


PROPOSITION; if

< Y r.

for each n s and |xl < V r . ,

n+i

i
i

then x can be expressed in the form (14)Q


The special case of Theorem 1 in which k. = 1 and m. = 0, for all i,
is apparently an old result first proved by Kakeya [7] (cf. [2]). Generalizations
of the n-scale (radix n) representation of positive integers which are analogous to the theory presented here have been developed by Alder [l] and Brown
[3-5].

REFERENCES
1.' Hs Le Alder, "The Number System in More General Scales, ! ? Math,' Mag c ,
35(1962)145-151 e
2." C. R Banerjee and B. K. Lahiri, ffQn Subseries of Divergent Series," Amer,
Math. Monthly, 71 (1964) 767-8 a

1966]

FOR THE REAL NUMBERS

201

3.

J . L 0 Brown, J r . , !f Note on Complete Sequences of I n t e g e r s , " A m e r . Math.


Monthly, 68 (1961) 5 5 7 - 560/
~~

4.

_,"A Generalization of Semi-Completeness for Integer Sequences," Fibonacci Quarterly, 1 (1963) 3 - 1 5 .

, f 'Generalized Bases for the Integers s f ?


Amer, Math.
Monthly, 71 (1964) 973-980.
6 J, A. Fridy, f?A Generalization of N-Scale Number Representation,"
Amer. Math. Monthly, to appear.
5.

7. S. Kakeya, "On the Partial Sums of an Infinite Series, f f Set


Tohoku Imp. U. (1), 3 (1914) 1 5 9 - 1 6 3 .

Reports

8. N.. S. Mendelsohn, "Representations of Positive Real Numbers by Infinite


Sequences of Integers," Trans. Royal Soc. Canada (3 Ser.) Sec. Ill, 47
(1952)45-55.
, ^ ^
ANNOUNCEMENT
The Journal of Combinatorial Theory will be devoted to the publication of original
mathematical articles dealing with theoretical and physical aspects of the study
of finite and discrete structures In all branches of science,, It will fully explore
the long-neglected field of combinatorial analysisa field which has only now
begun to realize its potential. The Journal will include the efforts of scientists,
working in many independent disciplines, who have discovered a common link
through combinatorics. Mathematicians, biologists, physicists, and engineers
will find this Journal a central reference for the combinatorial methods which
are increasingly important to many of their main problems. For furtherinformation, write Gian-Carlo Rota, The Rockefeller University, N. Y. , N. Y. , 10021.
VOLUME 1, 1966 (Issued quarterly beginning Spring 1966), $15.00
Add $1.00 for postage outside U. S A.
Personal Subscription $9.00
(Not for institutional use)
Add $1.00 for postage outside U. Se A.
Sample copies of the Journal are available upon request.

CONTENTS OF VOLUME 1, NUMBER 1:


George Polya, A Note of Welcome
Hans Rademacher, Robert Dickson, and Morris Plotkin, A Packing Problem
for Parallelepipeds
W. T. Tutte, On the Algebraic Theory of Graph Colourings
Cedric A. B. Smith, Graphs and Composite Games
John Rlordan, The Number of Faces of SImplicial Polytopes
R. C. Bose and R. C. Burton, A Characterization of Flat Spaces in a Finite
Geometry and the Uniqueness of the Hamming and the MacDonald Codes
Michael E. Fischer, On Hearing the Shape of a Drum
Henry H. Crapo. The Mobius Function of a Lattice
Jane DiPaola, Block Designs and Graph Theory
SHORT NOTES:
P. Erdos, On the Construction of Certain Graphs
Daniel J. Kleitman, Families of Non-Disjoint Subsets
RESEARCH PROBLEMS
Published by the Academic P r e s s .

ON A PARTIAL DIFFERENCE EQUATION OF L CARUTZ


W . Jentsch, University o f H a l l e / S . ,

Germany

Translated by: P . F , Byrd and MonTka Aumann, San Jose State C o l l e g e , San Jose, C a l i f .

SUMMARY
Eine von L CARLITZ behandelte partielle Differenzengleichung zweiter
Ordnung, die mit den FIBONACCI-Zahlen in Beziehung steht, wird mit Hilfe
einer algebraisch begriindeten, zweidimensionalen Operatorenrechnung gelost.
Die sich hierbei ergebende Losung ist allgemeiner als diejenigevon L. CARLITZ.

In an article [1] by L. Carlitz, a solution of the equation


(*)

mn ~ u m-i,n ~ u m,n-i ~ u m-2,n

+ 3u

m - i , n - i ~ u m,n~2 =
(m,n ^ 2, integral)

was given with the aid of a power series expansion related to the Fibonacci
numbers. Although the solution contains only three arbitrary constants (viz# ,
u00s u 01 , and u 1 0 ), it is called a "general solution" a terminology which
appears justified only if, besides equation (1), the following secondary conditions, not mentioned in [ 1 ] , are also imposed:
(2)

u u - u 01 - u10 + 3u 00 = 0 ,

(3)

u 0 n - u ^ ^ - u0jI12 = 0

(4)

u m 0 - um_1?0 - um2>0 = 0

(5)

u i n - uon - u t ^ i + S u o ^ i - uljn_2 = 0

(6)

for n > 2 ,
for m => 2 ,
for n > 2 ,

u m l - u m _ 1 ? 1 - u m o - u m _ 2 y l + 3 u m l j 0 = 0 for m > 2 .

The conclusion (1.4) from [1] is valid only under the assumptions (2) to (6).
From (2), u.n is fixed, and from (3) to (6) the initial values u 0 n , u l n

and

u m 0 , u m l are uniquely determined for n,m ^ 20 The general solution of (3),


for instance, is
202

203

ON A PARTIAL DIFFERENCE EQUATION

(7)

u o n - u01 F n + u00 Fn_t

[Oct.]':

for n > 0 ,

where
(8)

n _ n
F n = ^-zp

__
1
1
= ^ (1 + V5) , iS = |- (1 - NTS)

with

(n integral).

One can easily verify that the solution (5.4) given by LB Carlitz in [1] reduces
to equation (7) for m = 08
The general solution of (1) without secondary conditions thus contains two
pairs of arbitrary functions of only one of the two integral variables m and n.
We now wish to determine this solution with the aid of the

?f

two-dimensional,

discrete operational calculus" developed in [ 2 ] .


According to the fundamental idea of Je Mikusiriski (see perhaps [3]), the
usual addition and the two dimensional Cauchy product
m,n
(9)

mn * b mn ~

/ivbm-jU n-v

as

multiplication

/x ,v =0
are introduced in the set of complex-valued functions of two nonnegative integral
variables, and the quotient field Q2 belonging to the integral domain D2 a r i s ing from this means is considered.

In order to conform with the relations in

[ 2 ] , we make an index shift in (1) and write


(l f )

DfUum) = u m+2j n+2 ~ um+i,n-H2 ~ um+25n+t ~ um,n+2


- um+2?n

3u

m+i,n+i
(m>n -

After application of the difference theorem from [ 2 ] ,

m+k,n+i = P ^ m n

- Q* 2

nKn - p

q^umX +
k

~* ^
2
2
/c=o M
(u m n eD 2 ; u m , u.

lr.it M
pK V % xX
*

initial values; k? natural numbers;p,-.qinverses to shift

operators in Q 2 ), one obtains the operator representation

ON A PARTIAL D I F F E R E N C E EQUATION

204

(10)

[Oct.

= h(p,q,m ? n)
g(p,q)

w h e r e the n u m e r a t o r i s
h = ^ n h t + r m h 2 + /3 n h 3 + <5mh4 + or0h5 + ^ h 6 + /50h7 + fthg ,
a s one can e a s i l y verify with the polynomials
hi(p,q) = P 2 q 2 ~ pq 2 ~ P 2 q + 3pq - p 2 , h 2 = h t (q,p)
h 3 (p,q) = pq

- pq - P,

h 6 (p,q) = - p 2 q + p q ,

2 2

h 4 = h 3 (q,p),

h 5 = -p q

+ pq 2 + p 2 q - 3 p q ,

h 7 = h 6 (q,p) , h 8 = -pq

and the coefficients, the given initial v a l u e s ,


(11)

an

= uon , y m

o = ^o > ft)

= u m o , /5 n = u m , 6m
yu

o Pi = t

= uml

with

The d e n o m i n a t o r , a polynomial of d e g r e e 4 in p,q is


g(P,q) = P 2 q 2 - Pq 2 - P 2 q + 3pq - p 2 - q2 = g 1 (p,q)g 2 (p,q)
with
gi = pq - ap - ]Sq

and

g 2 = pq - /3p - aq

w h e r e a and ft have the v a l u e s given in (8).

As c a n be i m m e d i a t e l y p r o v e d ,

~~~x D 9
2
g(p,q)
holds for i = 1, , 8; and t h e s e t e r m s a r e indeed functions of the Fibonacci
numbers

F, .

If c o n s i d e r a t i o n s for the o p e r a t o r
p2q2
g(p,q)

a r e indicated, the calculation for the r e m a i n i n g m e m b e r s of u m n


easily.

If one conceives

then follows

1966]

OF L. CAELITZ

205

pq2
g(p?q)
as a (proper fractional) rational operator of p alone, then there results, by
decomposition into partial fractions,
p2q2
g(P
g(P,q)

i
/ o-pq _ /fog j
g__
a - p \g 2 (p,q)
gi(p,q)/ q - 1

and on account of the obvious relations

g2(p,q)

m )

'

g l (p,q)

\ m )

a P

and of the meaning of q/(q - 1) as a "partial summation operator"


-a b
=
q - 1 mn

2 b

m^ '

it follows that
gg(p,q)
(pW

-c --

/5 kk t= 0
P

I m /L^

from which, on account of a ^ k = (_i)k (^ integral), of definition (8), of the


symmetry of g(p,q), and with the notation G m n for (p 2 q 2 )/g , there

finally

results
a2)

g & = G - = <W = (-D k ( m k +k ) Fm+1_k =


m

s2

k=o

(-l)k(nkk)Fn+1_k

for m,n s

o.

With the aid of (12) the operators h j / g (i = l , - . . , 8 ) - can now be immediately

206

ON A PARTIAL DIFFERENCE EQUATION

[Oct.

represented as functions from D 2 . In order to simplify the notation, we define;


G m n = 0 in case an index is negative; according to (12) this is also achieved by
stipulating the following:
(13)

2 ak = 0 , I

1= 0

for < 0 .

(l/p 2 ) G m n = Gm^2>^.

(With this agreement,

? example,,.holds,for all m,n

Therewith we obtain, after easy calculation from (10),


(14)

u m n = ^ n ( l + G m _ 2$li ) + 7 m * (1 + G m ? n _ 2 )
+

0n" ( G m-i,n ~ G m-i,n~i ~ G m-i,n-2)

+ 6 m - (Gxa,n-i ~ G m - i , n - i " G m-2,n-i) - o G mn


+ {a0 - /30)Gm-i3,h + (o " ^i) G m f n-i "*" (3<*o - 0 ~ a\ + ^i) G ixi~i,n-i
for all m,n 0.
(In this the multiplication symbol means multiplication in D2 and the summand
1 is the identity element of D 2 .) If we finally use
G

m - i , n ~ G m~i,n-i " ('

D n (r)vn

and correspondingly
Gm,n-i - G m -i,n-i = ( - l ) 1 1 1 ^ " 1 ) F n . m

for m,n -

0 ,

and carry out the multiplication in D2 then after simple transformations


m,n ^ 0 we obtain from (14)
n
<15*

mn

+ y

V^/n-^ + E

jx.,n-2ym-ji

ii

ill

m
~J

m
x

ft

, i x m / m + n\
O(-D ( m j

n-jbt m-ju

JLd

/A-2fn-i m-/i

n / i x n / m + n - l \
F n _ m + 1 - /30(-D y
j Fm_n
n

for

1966]

OF L. CARLITZ

201

We verify that (15) satisfies equation (1 1 ) ? however, we only indicate the calculation; to begin with, D(orn) = -%+2 holds and D(7 m ) = -7 m +2
Furthermore,

2nGm-2^B^

n+g
= 2

%+2,D(Gm^2)

^ - ^ [ G ^ - G ^ ^ - G ^ J

+ a , JG
- G 4 -G -G
+ 3G
-l
n+iL mi
m - i , i mo m-2,i
m-i,o J
= Q

W2

for, it is true that G m 0 - G m . 1 ? 0 - G m . 2 ? 0 = { J r m > 0 '


G

m t ~ G m - i , t - G mo - G m-2,t

+ 3G

m-i 5 o

=:

for all m > 0

0 ,

and
D G

( HI-2SP-2)

for m ^

2 ,

k+i
as one recognizes after some calculation with the aid of (12) and F ^ = (-1)
F,
(k integral) or as one can read off directly from the fact that G m n in D2 i s
inverse to

V o o ...
by (9).
Analogously one completes the verification. By appropriate calculation one r e c ognizes that the initial conditions (11) are satisfied by (15). Since 60 = at, and
because of definition (13) and of the validity of the relation (3) for F n ? there
results for m = 0 S n ^ 0 , for instance,
u

on = < V o + GD,n-2 o

("'o^n'-^i^-^'o'K^'

208

ON A PARTIAL DIFFERENCE EQUATION


OF L. CARLITZ

TOct. 1966 1

REFERENCES
1. L. Carlitz, A partial difference equation related to the Fibonacci numbers,
Fibonacci Quarterly, Vol. 2, No. 3, pp. 185-196 (1964).
2. W. Jentsch, Operatorenrechnungfur Funktionenzweier diskreter Variabler,
Wiss. Zeitsehr. Univ. Halle, XIV, 4, pp# 311-318 (1965).
3. J. Mikusinski, Sur less fondements du calcul ope'ratoire, Studia Math. 11,
pp. 41-70 (1950).
* *

RECURRING SEQUENCES
Review of Book by Dov Jarden
By Brother Alfred Brousseau
For some time the volume, Recurring Sequences, by Dov Jarden has been
unavailable, but now a printing has been made of a revised version*

The new

book contains articles published by the author on Fibonacci numbers and r e lated matters in Riveon Lematematika and other publications.

A number of

these articles were originally in Hebrew and hence unavailable to the general
reading public. This volume now enables the reader to become acquainted with
this extensive material (some thirty articles) In convenient form.
In addition, there is a list of Fibonacci and Lucas numbers as well as
their known factorizations up to the 385th number In each case. Many new r e sults in this section are the work of John Brillhart of the University of San
Francisco and the University of California*
There is likewise, a Fibonacci bibliography which has been extended to
include articles to the year 1962a
This valuable reference for Fibonacci fanciers is now available through
the Fibonacci Association for the price of $6.00. All requests for the volume
should be sent to Brother Alfred Brousseau, Managing Editor, St.

Mary f s

College, Calif. , 94575.


*

The Fibonacci Association invites Educational Institutions to apply for Academic


Membership in the Association. The minimum subscription fee is $25 annually.
(Academic Members will receive two copies of each issue and will have their
names listed in the Journal.

ENUMERATION OF PARTITIONS SUBJECT TO LIMITATIONS ON SIZE OF MEMBERS


Daniel C . Fielder, Georgia Institute o f Technology, A t l a n t a , Georgia

1.' INTRODUCTION
In a previous work [1], it was shown that'the partition enumeration*
P(n[p| < n + p - 1) is given by

(II

P(n|p| < n + p - 1) =

"n - p + 2

+ E

n - p + 2 - w.
P*l,

1=1

(la)

P(n|p|

n + p - 1) = 1

P = l.

The w. are the sums of each partition in the set of partitions described by
PV(>3,<n - pj>l,< [(n-p)/3])^3 5 <pK

It was stated in [1] that the summa-

tion term of (1) is zero for those values of p and (n - p) for which PV(^3 9
< n - p | ^ l ? ^ [(n-p)/3}j^3,^p) does not e x i s t
brief description of nomenclature.)

(See the footnote below for

For n - p < 3 and/or p = 2, w. = 09

O n e raison d f etre for (1) is the adaptability of w. to digital computation,


P(njp!^n - p + 1) is a basic enumeration form which is extremely useful
in evaluating more restrictive enumerations [ 2 ] , PV(n|p[<n - p + 1) shares
this versatility in that sets of many other partition types can be constructed by
operations on the members of the partitions of the basic set When PV(n)p|^i
- p + 1) is under consideration, it is convenient to arrange the p members of
a partition so that
(2)

*p

p-i -

p~2

P(n)pj<q) is the enumeration of the partitions of n into exactly p members,


no member of which is greater than q The appended notation PV(n|p|^ q)
is the actual set of such partitions* The use of ^ and/or ^ symbols with n,
p , o r q defines lower limits and/or upper limits of the quantity modified. Note
that [ ] (except for obvious reference use) is used with realnumbers to indicate
the greatest integer less than or equal to the number bracketed,
209

210

[Oct.

ENUMERATION OF PARTITIONS

where a,

is an individual partition member. The arrangement of (2) leads to

an initial partition of PV(n|p|^n - p + 1) as


ap

(3)

p-i

1 1

p-2
1

n - p + 1

One method [3] of generating successive partitions of PV(n|p|<n - p + 1) starts


with (3) and successively increases a2 by 1 and decreases a.t by 1 until (2)
is just barely satisfied.

New members,

a , a p ^ , - , a 2 a i are chosen

exhaustively, and the increase a 2 decrease a^ process is repeated.


Based on the above brief background, it is possible to consider the following enumeration extensions to (1):
(a)

P(n|p|>s)

No member less than s, where s is a positive integer


such that s ^ n - p + 1 .

(b)

P(n|p|<r) .

No member greater than r where r is a positive integer


such that r ^ n - p + 1 .

(c)

P(n|p|^s, ^ r ) .

No member less than s, or greater than r


2. ENUMERATION OF P(n|p|>s)

There exists one member of a partition in the set PV(njpjs)


at least as large as any member of any partition in the set.
be q

which can readily be found as


s

(4)

= n - s(p - 1) .

This implies that for any a ls


(5)

n - ps^ + s ^ a t ^ s

from which a necessary condition of P(njp|^s) is seen to be

(6)

(?)

which is

Let this member

1966]

SUBJECT TO LIMITATIONS ON SIZE OF MEMBERS

211

The initial partition of PV(n|p|^s) is


(7)

s, s, s, - . . , s , n - s(p - 1)

If s - 1 is subtracted from each member of (7), the result is a modified initial partition
(8)

1, 1, 1,

aeo

, 1, n - sp + 1 .

The complete enumeration for a partition set starting with (8) i s , according to
(1), P(n ? |p|<n f - :p + 1), where
nf = n - sp + p .

(9)

Because the aA and a2 members of the initial partitions (7) and (8) differ by
the same integer (n - sp) and because each a^ of each partition developed
from (7) is (s - 1) greater than the corresponding a^ of the corresponding
partition developed from (8), there are exactly as many partitions developable
from the start of (7) as there are from (8). Hence, P(n|p|^s) appears in the
form of (1) as
(10)

P(n|p|^s) = P(nf|p|Mnf - p + 1) .

As a simple example, consider P(is|6|^2).

For this case, n = 15, and nT =

9.' It is seen below that P(15J6|^2) = P(9J6|<4) = 3e


PV(15[6l^2)
2,2,2,2,2,5
2,2,2,2,3,4
2,2,2,3,3,3
3e

PV(9[6l<4)
1,1,1,1,1,4
1,1,1,1,2,3
1,1,1,2,2,2

ENUMERATION OF P(n|p|<r)

The partitions of the set PV(>3,^n - p]>l,< [(n - p)/3] ] ^ 3 <p) canbe
arranged in columns according to the number of members in a partition.
is illustrated in Table 1 for n = 16, p = 58

This

212

ENUMERATION OF PARTITIONS

[Oct.

3,3

3,3,3

3,4

3,3,4

3,5

3,3,5

4,4

3,4,4

PV(> 3 , < l l | ^ 1, < 3 | > 3 5 <5)

4,5
5,5

Table 1
The sum of members of each partition is equal to a w. for use in (1). The use
of the index i can be extended somewhat to allow it to designate the column
from which the summed partition was taken. Although w. might stand for any
of several sums, no loss in generality results thereby since all of these sums
must eventually be considered 0 To account for the non-summation term i n ( l ) ,
a zero

column with a lone zero entry is added to indicate that an added w0 =

0. Table 2 shows values of w. for n =. 16, p = 5.

10

11

11

9
10

Table 2 Values of w.
If, as the w. f s are successively selected for enumerating

P(n|p|^n-p

+ 1) in (1), a simultaneous generation of the partitions in the set PV(nlp|^n p + 1) is made

(by the increase a 2 decrease &t method, for example) there

would result subsets of PV(n|p|^n - p + 1) each having [(n- p + 2 - wi)/2]


partitions of n0 For i = 0, the subset can easily be constructed,,

It is seen

that the a2 and a3 members of the initial partition must necessarily be one.
For i = 1, the a2 and a3 members of the initial partition assume the least

SUBJECT TO LIMITATIONS ON SIZE OF MEMBERS

1966

possible value two since i = 0 has accounted for the value one9

213

It can be

argued in this fashion that the a2 and a3 members of an initial partition in a


subset must be (i + l ) . The a4 member of the initial partition of the subset
would not generally be known in advance.

However, this member is certainly

not less than any member of any partition in the subset.

Set dj be the a1

member of the initial partition corresponding to the particular w-. If bj are


the number of partitions In the subset, the bracketed terms of (1) limit the
possibilities of bj to either
(11)

n - p + '1

2b t

or

n - p + 1 - w. = 2b|

(11a)

The arrangement of the subset of b^ partitions is


ap
X

(12)

ap_i
X

aj

Sin

1+1

i+1

d| ** (Initial Partition)

i
partitions
i+1

I+bj

di-bi+1

From (12), it can be deduced that either


(13)

d{ = 2b. - 1 + i

or

(13a)

d. = 2b. + i
i

Comparison of (13) with (11) and (13a) with (11a) yields the desired

(14)

dj = (n - p + 1 - wj + i)

ENUMERATION OF PARTITIONS

214

[Oct.

An i l l u s t r a t i o n i s given in Table 3 for c o n s t r u c t i o n of P V ( 1 6 | 5 | ^ 1 2 ) , c o n s i s t e n t


with the w. from

P V { ^ 3 , ^ l l | ^ l , < 3 | ^ 3 , <5) a s a r r a n g e d in T a b l e s 1 and 2.

1,1,1,1,1,12

1 , 1 , 2 , 2 , 10

1,1,3,3,8

1,1,4,4,6 1

1,1,1,1,2,11

1,1,2,3,

1,1,3,4,7

1,1,4,4,5

1,1,1,1,3,10

1,1,2,4,

1,1,3,5,6

1,1,1,1,4,

1,1,2,5,

1.1.1.1.5,

1,1,2,6,

1.1.1.1.6,

7
1,2,2,2,

1,2,3,3,7

1,2,2,3,

1,2,3,4,6

1,2,2,4,

1,2,3,5,5

1,2,2,5,

PV(16| 5|^12)

2,2,2,2, 8

1,3,3,3,6

2,2,2,3,

1,3,3,4,5

2,2,2,4,

2,2,2,5,

5
2,2,3,3,6

1,2,4,4,5

1,3,4,4,4

2,2,4,4,4

2,2,3,4,5
Z , O , Oj O , O

2,3,3,4,4
O,0,0,0,4t

Table 3 PV(16|5|<12)
Table 4 shows bj c o r r e s p o n d i n g to wi
i

1
Table 4

of Table 1 for

1,

5
4
4

3
3
2
2
2
1

2
1
1
1

2 b.l = P(16J5j<12) = 37
i

P(16|5|^12) = b.
i=o

1966]

SUBJECT TO LIMITATIONS ON SIZE OF MEMBERS


Let the b. for P(n|p|<r) be b i r

It follows that for

215

P(nlp|<r) each

b j r can have no more (and will possibly have less) than bf partitions.

The

non-negative integer by which b j r is less than b r can be observed by comparing r with the entries in the SL1 column of (12),

This leads immediately

to

"n - p + 2"
(15)

P(n|p|<r) =

<*o

n - p + 2 - w.

Ebi

"') = i

where
0

(16)

(r>d.)

d.-r

(d. > r ^ (d. - b.


x

n - p + 2 - w.
2

'

~_

(r < (d. - b. + 1))

Table 5 serves to illustrate (15) for n = 1 6 , p = 5, r = 7ffi


i

1 2

bir

3
2

Table 5 E

ir

P(16J5J^7) = 23

4 ENUMERATION OF P(n|pl>s,<r)
The combination of the previous two methods leads quickly to the desired
enumeration. Reference to (10) reveals a P(n ? |p|^n f - p + 1) for which every
member of each partition of PV(n f ]p)^n f - p + 1) is (s - 1) less than the corresponding member of the appropriate counterpart in PV(n|p|^s). If the desired r is depressed to r f where

216

ENUMERATION OF PARTITIONS SUBJECT


TO LIMITATIONS ON SIZE OF MEMBERS
r ! = r - (s - 1)

(17)

[Oct. 1966]

the enumeration P(n}|p|^s,^r) is equal to P(n f |p|^r f ) .

REFERENCES
1. D. C0 Fielder j "Partition Enumeration by Means of Simpler Partitions,"
Fibonacci Quarterly, Vol. 2, No. 2, pp 115 118, 1964.
2. G8 Chrystal, Textbook of Algebra, Vol. 2 (Reprint) Chelsea Publishing Co0 ,
New York, N. Y. , pp 5 5 5 - 5 6 5 , 1952.
3. D. C. Fielder, !?A Combinatorial-Digital Computation of a Network P a r a m eter," IRE Trans, on Circuit Theory, Vol. PGCT-8, No. 3, pp 202209,
Sept. 1961.

DID YOU KNOW?


Prof. D.'E. Knuth of California Institute of Technology is working on a 3 volume book, The Analysis of Algorithms, which has 39 exercises at the end
of the section which introduces the Fibonacci Sequence. However, the Fibonacci
Sequence occurs in many different places, both as an operational tool, or to
serve as examples of good sequences and also bad sequences. He reports that
there are at least 12 different algorithms directly or indirectly connected with
the Fibonacci Sequence.

In the age of computers, the Fibonacci Sequence is

coming of age in many ways e

This book will be a most welcome addition to the

growing list of Fibonacci related books and articles.


*
Prof. C. T. Long of Washington State University has written a very nice
book, Elementary Introduction to Number Theory, 1965, Heath, Boston. It contains a good discussion of the Fibonacci Numbers in Chapter One and several
Fibonacci Problems in Chapters I and II.

ON THE DIVISIBILITY PROPERTIES OF FIBONACCI NUMBERS


John H s H a l t o n , University o f Colorado, Boulder, Colorado

1.

INTRODUCTION

The Fibonacci s e q u e n c e i s defined by the r e c u r r e n c e r e l a t i o n

(1)

Fn+2 = Fn+1 + F n

t o g e t h e r with the p a r t i c u l a r v a l u e s

F 0 = 0,

Ft

= 1

whence
F 2 = 1, F s = 2, F 4 = 3 , F 5 = 5, F 6 = 8 = 2\

F 7 = 13

F 8 = 21 = 3 " 7, F 9 = 34 = 2 17, F 1 0 = 55 = 5 1 1 ,

and, in p a r t i c u l a r ,
(2)

F 1 2 .= 1 4 4 = 2 4 3 2 , F 1 4 = 377 = 13=29, F 1 5 = 610 = 2 - 5 6 1 ,


F 1 8 = 2584 = 2 3 -17 . 19, F 2 0 = 6765 = 3 5 11 4 1 ,
F 2 1 = 10946 = 2 1 3 - 4 2 1 , F 2 4 = 46368 = 2 . 5 - 3 2 - 7 - 2 3 ,
F 2 5 = 75025 = 52 3001, F 2 8 = 317811 = 3 - 1 3 29 281 ,
F 3 0 = 832040 = 2 3 5 11 31 6 1 , F 3 5 = 9227465 = 5 13 141961 ,
F 3 6 = 14930352 = 24. 3 3 -17 1 9 - 1 0 7 , F 4 2 = 267914296 = 2 3 - 1 3 - 2 9 - 2 1 1 - 4 2 1
F 7 0 = 190392490709135 = 5 11 13 - 29 7.1 911 141961

In this p a p e r , we shall be c o n c e r n e d with the s u b - s e q u e n c e of Fibonacci


n u m b e r s which a r e divisible by p o w e r s of a given integer,,

We shall a l s o be

i n t e r e s t e d in the a s s o c i a t e d p r o b l e m of the p e r i o d i c n a t u r e of the s e q u e n c e of


r e m a i n d e r s , when the Fibonacci n u m b e r s a r e divided by a given i n t e g e r .
217

218

ON THE DIVISIBILITY PROPERTIES

[Oct.

The Fibonacci sequence is defined for all integer values of the index n0
However, the well-known identity

(3)
%
'

=
-n

(-l^F
'
n

shows that negative indices add nothing to the divisibility properties of the
Fibonacci numbers,,

We shall consequently simplify our discussion,

without

loss of generality, by imposing the restriction that n ^ 0.


Of the many papers dealing with our problem, perhaps the most useful
are those of Carmichael [1] , Robinson [5] , Vinson [6] , and Wall [7]; and the
reader can find many additional references in these.

Most of the other papers

in the field give either less complete results, or give them for more general
sequences.
We shall make use, in what follows, of the well-known identities:*

^{(H^)"-(4^)"} :

(4)

(5)

(6)

(7

>

kn + r

n " Fn_iFn+1 =

(-if1

VhFk-hF
I n n-i

if k > 0
r+h'

and since F 0 = 0,

(8)

F ^ = Fn2-fL )Fn lF n-i F h


h= v n /
h=l

*See, for example, equations (6), (3), (5), (67), and (34), in my earlier t paper
[ 3] Equation (5) above follows from (4) by the binomial theorem.

1966]

OF FIBONACCI NUMBERS

219

Also

n=c>L\
and

(10>

p divides I

J if p is prime and 0 < s < p ,

::)

and Fermat ? s theorem, that


(11)

nr

= 1 (mod p) if p is prime and (m 9 p) = 1

As is customary, we use (A 9 B,C, o o ) to represent the greatest common factor of integers A, B5 C, e , and [A, B s C, ] to represent their least
common multiple. We have

n^P""1) = ( m / p ) ( m o d

(12)

p)

where p is an odd prime and (m/p) denotes the Legendre index, which is 1
if (m,p) = 1, and 0 otherwise.
Each writer seems to have invented his own notation0

I shall adopt the

following, which comes closest to that of Robinson in [ 5 ] .


Definition 1, The least positive index a such that F
1

(13)

(that i s , F

is divisible by

= 0 (mod m )) will be written


ar(m,n) = aim1,1)

= a(mn)

This is variously called the "rank of apparition" (why not "appearance"?) of


mn,

or the "restricted period" of .the Fibonacci sequence modulo mn


Definition 2. The least positive index p. such that both F =r 0

Fjjtf-i = 1 (mod m11) will be written

and

220

ON THE DIVISIBILITY PROPERTIES


M(m,n) = M(m n ,l) = M (m n )

(14)

[Oct.

This notation follows Carmichael [2], who named \x the "characteristic number"
of the Fibonacci sequence modulo m n c It is also called the "period" of the s e quence modulo m n .
Definition 3. I shall write

(15)

m.

jLi(m,n)/Qr(m,n) - /3(m,n) = /3(mn, 1) - /3(mn)

J
Definition__JL
40 The greatest
integer
v such that F or(m,n)
,
. i s divisible by
&
&
will be written

p(m s n) - i>(m n ,l) = v(mP)

(16)

It is then clear that


(17) or(m,n) = oj(m,n + 1) = = ar(m, p(m,n)) < a(m,p(m9ii)

+ 1)

or, equivalently,
(18)

v(m, i/(m,n)) = v(m,n)

Definition 5Q I shall call the sequence


(19)

a(m, i) >

f f (m,

2) ' " * ' Faf(m, n)' "

'

the divisibility sequence of m


2 PRELIMINARIES
We shall need a number of preliminary results, whose proofs will be outlined for completeness,,
Lemma 1, F , F

, and F

are always pairwise prime,

Clf f divides two of the numbers, it must divide the third, by (l) c

Thus, , by

1966]

OF FIBONACCI NUMBERS

221

induction along the sequence, using (1), we see that f must divide every F .
Thus, since Fj = 1, f = 1.]
Lemma 20 If n ^ 2, F

is a strictly increasing positive function of n0

[By (1), if F n ^ ^ 0 and F ^ * 1, F n + 1 >

FQ

2> 1.

By (2),

F 0 = 0 and

F1 = 1, whence the lemma follows by induction/]


Lemma 38 If n ^ 3
(20)
1fBy
J

<*(Fn) = n. .
Lemma 2, if n ^ 3, the least index m such that F ^ F is
.
'
'
m
n

nJj

Lemma 4
(2D

(Fm,Fn) = F(m>n)

[Let (m,n) = g and ( F m , F n ) = G0 There are integers x and y (not both


negative) such that xm + yn = g0 Suppose x ^ 0; then, by (7),
x

/
= I P I
g
Z-# \
h=o

\
) F h F X " h F M, = 0 x(mod G); ?,
m m-i yn+h

x n /

since G divides F

and F , and by
(8),
F divides F . Thus F
is
J W
J
m
n*
n
yn
g
divisible by Ge Again, by (8), F, ^ = 0 (mod F L Thus, since g divides
and F , and so G is divisible by
both m and n, F divides both F
Lemma 5a F

is divisible by F , if and only if either m is divisible

by n, or n = 20
[By Lemma 4, ( F m , F n ) = F n if and only if F (

^ = F n ; that i s , (m,n) =

n or n = 2B ]
Definition 6n The remainder when F

is divided by m will be written

F^ ' and will be called the residue of F modulo m0


n
n
(22)
v

F = F(m)(modm),
/5
n
n x

Clearly

0 ^ F(m) < m
n

222

ON THE DIVISIBILITY PROPERTIES

[Oct

Lemma 6C The sequence of residues F ^ m ' , modulo any integer m>


2, is periodic with period /x(m).' That is

(23)

or

(m)
n+kM(m)

(m)
n

F ,, , , = F x (mod m),
n+l^(m)
n
'
[The ordered pair of integers F^ , F ^ , ; can take at most m2 distinct valn
n+i
/ \
/ \
/ \
ues 0 Thus the m2 + 1 such consecutive pairs in F J , F\ ' , , F- 2_L
must have a duplication. By backward induction on the indices of two equal
pairs, using (1), we see that there must be a pair F 5 ' , F k + i equal to F J
= 0, F j m ' = 1, with 2 ^ k ^ m20 By definition, the least such k is

pt(fta),

The periodicity now follows from (1). ]


Lemma 7, For any integer m, we can find an F

divisible by m,

[For example, p = kju(m)9 for any integer k, by Lemma 6]


Lemma 8e F

is divisible by m if and only if n is divisible by a(m)0

[Since m is a factor of F ,

v;

if n is divisible by o?(m), F

is

divisible

by m, by Lemma 50 Let n = k&(m) + r , 0 < r < a(m), and let m

divide

J
F n .' Then, by
, , F r = F.n = 0 x(mod m).
Lemma *1,
J (7),
v / FQ/(m)-i
' Thus,s sinceby
F
F
= 1;
( cKm)> or(m)-i)
F. = 0 (mod m). Since r < a(m), which is minimal,
F = 0 ; whence r = 0 and n is divisible by a(m)e]

Lemma 9, For all integers m and r > s > 0, a(m,s) divides a(m,r)
r
s
[F
,
v is divisible by
and so by
L
J m
J m e The result follows from Lemma
#(m,r)
8.]
Lemma 10. M(m) is divisible by a(m)Q That i s , /3(m) is an integer,,

[Since F y ,

= F i m ' = 0, F"

is divisible by m0

The lemma follows

from Lemma 8. ]
Lemma 11, If p is an odd prime, then p divides only one of
F , and F

; namely, F m , where m = p - (5/p).'

[(p,2) = 10 Using (5), (10), and (11), we obtain that


1/

(24)

F p s 2P~
s=o

1966

OF FIBONACCI NUMBERS

223

Thus p divides F if and only if (5/p) = 0, by (12); that i s , when p = 5,


By (5), (9), (10), and (11),
4<P-i)
(25)
y

2F ^ = 2 F ^
p+i
p+i

(\2s+ 1/

\2s/)

and, by (1), (24), and (25),

(26)

2F

The lemma now follows.

s 1 - 5^P

l)

(mod p)

We may, note that all but the dependence on (5/p)

follows directly from (6), which yields that, if p + 5, by (11) and (24),
F p~i Fp+i
__, = Fp2 - 1 s 0 \(modp)
r?

*;

and from (1).]


Lemma 12, a(p) divides p - (5/p), if p is an odd prime; and if a(p)
is itself prime and p 4= 5, a(p) < p
[ The first part follows from Lemmas 8 and 11. Thus a(p) < p .+ 1. By Lemma 11, if p # 5 and a(p) is prime, since p 1 is not prime, a(p) ^ p - 2.]
Lemma 13, If

K
i

(27)

m = Pi p 2 p k

where the p. are distinct primes and the X. are positive integers, then

(28)

ar(m,n) = [^(p^nXi), aip2sn\2)f

, a<&>*>%)I

M(m,n) = ( X p ^ n ^ ) , ii(p2,n\2),

and
(29)

M(Pk 9 n %)]

224

ON THE DIVISIBILITY PROPERTIES

[By Lemma 8, F, is divisible by p.


n \ . ) . Thus F + is divisible by m
of(p.,n\. )c
F

Since a(m,n)

(mod p.

[Oct.

if and only if t is divisible by #(p.,

if and only if t is a multiple of all the

is minimal, (28) follows,,

By Lemma 6,

, =

) for every s if and only if t is a multiple of ^(p.,nX.) Thus,

by the Chinese remainder theorem, F

, = F (mod m ) for every s if and


s+t
s
only if t is a common multiple of the M(p.,n\.) 0 Since ju(m,n) i s the mini-

mal such t, (29) follows,]


Lemma 140 For any integers m and n,
or([m,n]) = [a(m), a(n)]
(3.0)

and
M([m,n]) - [iMm), M(n)]

[This follows from Lemma 13, by expanding m and n in prime factors, ]


Definition
Definitio] 7. The greatest integer n such that N is divisible by m
will be written
(31)

n = pot

and called the "potency" of N to base m, following Hc Guptac It is then clear


that, in particular,
(32)
x
'

v(m,n) = pot F .
.
v
v
' '
m of(m,n)

Lemma 15a P t m F N = n if and only if N is divisible by cv(m,n) but


not by a(m,n + l) c
[ This is an immediate consequence of Lemma 8 ]
Lemma 16c If k and n are positive integers, then ( F - ^ / F , F ) is
a factor of k
[By (8), F ^ / F = k F ^ 1 (mod F n ).' Thus, if ( F ^ / F F n ) = g, g divides
k-i
kF
. B y Lemma 1, (F ' F ) = 1; so g divides k j
J
&
n-i
' v n-i J n 7
'
J

1966]

OF FIBONACCI NUMBERS

225

Lemma 17, If k and n are both integers greater than one, then
F|Q! / F n is a strictly increasing function of n and of k.
[By (8), F ^ / F n = 2 h = 1 L p
and increases with F , F
n*

n~i*9

F ^ F ^ Every t e r m in the sum is positive,

and ke The result follows from Lemma 2 1


J

Of these results, those in Lemmas 1, 2, 4 7, 11, and 16 have been


known for a long time s

Lemmas 8 10 and 12 15 appear, or are implicit, in

the papers of Robinson [ 5 ] , Vinson [ 6 ] , and Wall [7][My a(m)9 /3(m),

ju(m)

are written a(m), /3(m), 5(m) by Robinson, and f(m), t(m), s(m) by Vinson,,
respectively; and Wall writes d(m), k(m) for my a(m)9 jLt(m)0]
30
Theorem l c

THE DIVISIBILITY SEQUENCE

If p is an odd prime and n ^ v(p), then


<*(p,n) - pn~v(p)<*(p)

(33)
(34)

v(p,n) = n

If p # 5,

(p,ar(p)) = 1; while
Qf(5,n) = 5 n

(35)
Further,
(36)

a(2) = 3, a(4) = 6 = a(S)

and if n ^ 3,
(37)

or(2,n) - 2 n 2a(2) = 2n""2 3


Proof. By Lemma 9, cKp,n) = k&(p,n - 1), for some integer k9 Write

F , , = p n A, F ,
. = p^B, F ,
. 4 =
s
ar(p,n)
^ ' or(p,n-i)
^
a(p,n-i)-i
Then, by (8),

226

ON THE DIVISIBILITY P R O P E R T I E S
/,v

(38)

if

)p (n " 1)(h - 1) B h C k - h F h .

pA = 2 (

T h u s , if EL > v(p) ^ 1,

[Oct.

since

(p, C) = 1, kB m u s t be divisible by p Hence,

v(p, n - 1) = n - 1, (p,B) = 1, whence p divides k

s o k,

is minimal,

BCr

(mod p) 0

whence
v(p),

k = pc

Now, by (10), s i n c e k > 2,

Since

a(p 5 n)

(38) y i e l d s that A =

Since the f a c t o r s on the r i g h t a r e p r i m e to p ,

v(p,n) = n0
(34)holds and

By (18),

v(p, v(p)) = y(p),

and

so is

A,

s o that, by induction, if n ^

ar(p,n) = p n ~ V H * ( p , v(p)).

By (17),

or(p, v(p = or(p),

yielding (33).
By L e m m a 12, a(p)

divides p - (5/p).

If p = 5, then, by (2), a (5) = 5,- v(5) = 1,


Finally, if p = 2 5
= 2 if

(p,a(p)) = 1

and, by (33), we get (35).

(38) still h o l d s , and we s e e , a s b e f o r e , that k = p

v(2, n - 1) = n - 1.

and -v(2,n) = n,

T h u s , if p 4 5,

Thus

2A - 2 n ~ 1 B 2 + 2BC,

a s b e f o r e , if n 3.

whence (2, A) = 1

By (2), we have (36), whence we obtain

(37) like (33).


Theorem^

w h e r e p i s p r i m e and p + 2 , and
n
if r 0 and (p,t) = 1; then pot F
= n + r 8 If p = 2 , t m i s an odd
p
p tm
multiple of 3 and F,
i s an odd multiple of 2 , while, If r > 1, pot 2 F
t m
2rtm
r + 2e
Proof c
4= 2,

If p o t . F

= n ^ 1,

We r e p e a t e d l y u s e L e m m a 15 and T h e o r e m 1. If n >: 1 and p

e i t h e r p i s odd and n ^ v(p),

or p = 2 and n > 3; whence, by (33)

o r (37),
(39)
Thus,

or(p,n+ r)
m = ka(p9n)

t h a t pot F

= p r a ( p , n)

for s o m e k p r i m e to p 0 Hence p r t m = t k o r ( p , n + r ) ,

so

= n + r 0 By (36), if p = 2, m and t m a r e divisible by 3


prtm
but not by 6, so that pot 2 F,
= 1, and s i m i l a r l y by (37), pot 2 F
= r + 2,
l m
..
^ 2rtm
if r ^ 1.
p

T h e o r e m s 1 and 2 have a fairly long history,,

L u c a s [4] (see p a g e s 209

210) p r o v e d the s i m p l e s t f o r m u l a (39) with r = 1, but failed to notice the anomaly

1966]

OF FIBONACCI NUMBERS

227

when p n = 20 Carmichael .[1] (see pages 40 42) proved Theorem 2 in full,*


using the theory of cyclotomic polynomials,, Both Lucas 1 and CarmichaePs r e sults apply to a more general sequence** than that defined by (1) and (2).
Robinson [5] proves Theorem 1, for odd primes only, by a matrix method0
Theorem 30 If pot F . = n ^ 1 , where p is prime and p =1= 2, and
if r ^ 0; then there is a strictly increasing sequence of pairwise prime integers i = i (m,p)(s = 0,1,2,?**), all prime to p , such that
s
s
F
= Pn+Thhaa
p m

(40)-

^T

Proof. When r = 0, we define F = p J0, . where (p,40) = 1. By


Theorem 2, if r ^ 1, there are integers A,B, and C, such that

p m

= P ^ A , F r _ t = p ^ - ^ B , F r_1
= C ,
p m
p m-i

and (p,A) = (p,B) = 1, while, by Lemma 1,

(41)

= B E(

(pB, C) = 1. Thus, by (8),

) P^^^HB^CP-^

where, as in the proof of Theorem 1, the sum on the right is an integer, since
n 2: l . Thus A is divisible by Ba If we write A = i B s

it is clear that

A = H0 j I , yielding (40). Further (41) gives us that

(42)

= PB

Z)(

) P^n+r-l)(h-l)-2Bh-2Cp-hFh + C^1 ,

* He has a misprint, making the greatest power of 2 too small by one.


**The-sequence is D = (op- - j3n)/(a - /3), where a + f$ and a/3 are mutually
prime integers. For F n , by (4), a = (l/2)(l + N/"5) and p'= (l/2)(l - N/TT).

[Oct

ON THE DIVISIBILITY PROPERTIES

228

where the sum is again an integer, since either p ^ 3 and n ^ 1, or p = 2


and n ^

3Q Thus it

= Cr

(modpB); so that, since C

is prime to p,

^ o , ^ , " ^ _ , so is- 4 . Again, since exceeds a positive integer multiple


of pB, we have that
<43)

*r > P V I ' - - V i
Corollary 10 If pot F

Vi

= n ^ 1 and p

r_s(psm,

(44)

>

p)

+ 2, and if r > s > 0, then

Vm,p)

and

0(PSm> P) = 4o(mP)JMmP) - V m P )

(45)

Corollary 20 If p o t 2 F m = 1 and r > 1, then

(46)

r+2
= 21 0 ( 2 m J 2 ) 1 ( 2 m J 2 ) ^ r

(2m,2)

2 m
Theorem 3, with its corollaries, contains a definition of (m,p) whens
ever pot F = n > 1 and p n 4 28 By analogy with (40), (44), (45) and (46),
we shall adopt the following definition for the remaining case,
Definition 8, If m = 3t where t is odd (so that, by Theorem 2, p o t 2 F m
= 1), the sequence (m, 2) is defined by
s
(47)

.0(m,2) = | F

(48)

l,(m,2) = 2* 0 (2m,2)/ 0 (m,2)

and
(49)

ym,2) = *

(2m,2) if s > 2

1966]

OF FIBONACCI NUMBERS

229

Corollary 3S Adopting Definition .8, we obtain equation (40) for every


prime p, and every positive integer m such that pot F = n > 1. In every
case, the numbers Jta = 0 (m,p)(s = 0,1,2',) are integers, all pairwise
o

prime, and all but fL^m^) are always prime to p0 If m is an odd multiple
of 3,4^(211,2) is an odd multiple of 2; in every other case, J^(m,p) is prime
to p0
Proof0

If p

+2,

the corollary coincides with Theorem 30

If p n = 2

(that is, m is an odd multiple of 3, by (36)) and r ^ 1, Corollary 2 and Definition 8 (equations (46), (48), and (49)) show that equation (40) holds, with
^ 0 (m,2)j^(m,2), J02(m,2),j3(m,2),
Theorem 3. Finally, when p

all pairwise prime odd integers, by

= 2 and r = 0, we get (40) from the definition

(47)j and, by Theorem 2, J0o(m,2) is an odd integer.


Further, by (8), F 2 m = F m ( F m + 2F m _ 1 ), which yields through (40) that
i(m,2) = J20(m,2) + F m - l 9

Since %9Fm.t)

= (l 1 } F m -i) = 1 (by Lemma 4),

and both 0 and F m . j are odd, we see that ^(111,2) is even and prime to
|>0(m,2). Finally, since ji^
is odd, it must be an odd multiple of 2e
Theorem 4C Let P = {Pi,P2," # ,P k } be a set of k distinct primes,
Then P contains all the prime factors of F

, F , * , Fn7
PK
Pi P2

only if

k = 1 and P = {2} or {5} ,


(50)

k = 2 and P = {2, 3} or {2, 5} ,


or
k = 3 and P - -T2, 3, 5}

Proof0

Let k = 1. Then F

can have no prime factor other than p1#

By (2), Lemma 2, and Lemma 11, the only possible values of p 1 are 2 and 5
Let k ^ 2,

and first suppose that 2 $ P e

By Lemma 4, if I + ]\,

(F p ., Fp.) = 1, so that no prime factor is common to two of the Fp^; and by


Lemma 2, since every p. ^ 3,

every

Fp. has at least one prime factor,

Thus every F p . has exactly one prime factor e

Let us now renumber the p.,

if necessary, so that p 1 is the least prime in P not equal to 5, and

(51).

F^ = p^ 2 , F
= p3r3? , F
- p.1
3
Pi
P2
Pi-i
1

230

ON THE DIVISIBILITY PROPERTIES

[Oct

where each r j ^ 1, This can always be done, and, since p t 4= 5, inductively


P2?P3?000 + 5s> and no P i ! = p| a Finally, by Lemma 8S each pi 4 = ar(pi)
and so, by Lemma 12, or(Pi) = P i - t < Pi. Thus the sequence defined by (51)
cannot terminate, and this contradicts the finiteness of P,

Therefore

2eP

and we may write pt = 2a If the F p . (i = 2 , 3 , ,k) are all odd, the p^(i =
2,3, ,k) from a set of k - 1 distinct odd primes containing all the prime
factors of the corresponding" set of F p . # We have just shown that this can only
happen if k - 1 = 1 and p2 = 5, Suppose now that one of the F p . is even0
Then, by (2), we can write p 2 = 3, since F 3 = 2 If k = 2, this completes
the enumeration of possible cases 0 If k ^ 3 , then P3?P49 ? Pu form a set of
k - 2 distinct odd primes containing all the prime factors of the corresponding
set of Fp., because a(S) = 4, which is not prime, Again, we know that this
can only happen if k - 2 = 1 and p 3 = 5, This completes the proof0
Definition 9 If pot F N = n, and if either n ^ 1 and p = 5, or n >
v(p), we shall call p a multiple prime factor (mpf) of F N . If, on the contrary,
p + 5 and n = v(p), then p is a simple prime factor (spf) of F N .
Lemma 18, p is a multiple prime factor of F N if and only if it is a prime
factor of both F N and N, A prime factor of F N which is not multiple is, a
simple prime factor,
[This follows from Definition 9, Lemma 8, and Theorem 1]
Lemma 19, If k and n are positive integers and p is a multiple prime
factor, of F , it is also a multiple prime factor of F, . Conversely, if p is
a simple prime factor of F, , it is also a simple prime factor of F
[This follows from Lemmas 5 and 18.]
Theorem 5, F N has at least one simple prime factor, unless N = 1,2,
5, 6, or 12.
Proof.

Fj = F 2 = 1, so that these F N have no prime factors at all, and

so no spf, as stated.

Let N ^ 3, and let F N = m satisfy (27). By Lemma 2,

the set P of prime factors of F N is not empty.

If F N has only mpfs, by

Lemma.: 16, each p divides N; whence by Lemma 5, each F . divides m,


*

Pi

It follows that P contains all the prime factors of every F p . e This is the
situation dealt with in Theorem 4, and it can only occur in the five cases listed
in (50),
By (2), (50), Lemma 8, and Theorem. 1,. if F N has only mpfs, we see
that F N = 2 r 3 s 5t0

Further, r < 4; s < 2; t < 1; r t = 0; st = 0;

if

1966]

OF FIBONACCI NUMBERS

231

r = 0 then s = 0 and t = 1; if s = t = 0 then r = 3; if r s > 0 then r =


4 and s = 20 Thus F N = 5,8 9 or 144; whence N = 5, 6y or 12; and all
these cases are valid and stated in the theorem.
4m CARM1CHAEL?S THEOREM
By using the theory, of cyclotomic polynomials, Carmichael proved, for
the general sequence* D , a theorem which,, in our terminology, reads as
n
follows [Compare [1] , Theorem XXIII, pages 61 62e]
Carmichael f s Theorem. If N f 1, 2, 6, or 12, then there is a prime
p, such that N = #(p)0
We shall proceed to derive this theorem, for the Fibonacci sequence, by
the elementary considerations we have used so far.
(52)

N = q?*q?2...

Let

q *k

where the q. are distinct primes and the n^ ^ 1.

We shall write N,-v

for

any of the k integers N. = N/q*, and more generally N / M for any of the
k\
@
q ^ , with{i 1? i 2 ,** , %} a subset" (without r e p e h ] integers N/q. q.
tition) of {1, 2, % k}. We shall also write R, for the product of the
(, J
integers ~F N ..
Lemma 20. If N satisfies (52), then

h=i
[By repeated application of Lemma 4 we see that
< F NV %

"

Nih) =

<54>

*See footnote on page 227 above.

(N/qit, N / q ^ . - . - . N / q ^ ) =
F

N/

Vi2"' % "

rN

(h)

232

ON THE DIVISIBILITY PROPERTIES

[Oct.

so that Ru is the product of the, greatest common factors of all sets of h


numbers F M
N

. Let a prime factor p divide exactly s 1 of the F , ,


N

(D

; and

(D

let p 2 f p V , p m divide s 9 .s 3 ,* , s m of the F M , respectively; but let


(1)
no F
be divisible by p
Then k - s^ s 2 s m 1 and
pot [F

,F

( h1 ) ' ( h)

, , F N ]= ms Of the I ' 1 factors in R, ,


?

A if1 /

are

(54) shows that

respectively divisible by p, p 2 , *, p (Note

that J , 1 = 0 if s < h, and that the set of factors divisible by p

includes

those divisible by p 2 ? which include those divisible by p 3 s and so on)e

xyh-fh 1 ) + ( ? ) + -

( s if)

wh oe

t=i h=i

Thus

t=i

which implies (53)0]


It follows from Lemmas 5 and 20 that

F R2 R4 *
Q. - N
'N
R t R3 R 5 - - -

(55)

[ F ^ , F ^ , $ F N ^ ;

is a positive integer c [Carmichael [1] writes D M for my F N ?


N)

= /3^ QN(pr/i8) for my Qw

(56)

where

N) = q ? r V i - l j q ? 2 " 1 ^ - 1) ' " %k

in the Euler </>~functione ]

and F N (#,/3)

(% ~ D

1966]

OF FIBONACCI NUMBERS

233

By (55) and Theorem 2, if a prime p divides Q N , it is either a factor


of F N which is prime to every F
, or it also divides some F M . In
N
*(!>
(l)
the former case, by Lemma 8, since a(p) divides N, but no N m , necessarily N = of(p), and if N t 5, p 4= 5 and p is a spf of FN
case, by Theorem 29 p is a mpf of F

and pot Q

In the latter

= 1, except if N = 6

(when Q N = Q6 = F ^ / F 2 F 3 = 4.)
Lemma 21.' If N satisfies (52) and
(57)

Q N > q4q2 - qfe

then there is a prime p such that N = aip).


[As explained above, if N = 6, Q N = 4 < 2 3, so this case does not arise*
Thus pot Q = 0 or 1 and Q / q ^ '" \
cannot be divisible by any q.
Thus if this quotient exceeds one, Q

must be divisible by some prime other

than the qj, and such a prime p has N = or(p).]


Lemma 22. If N satisfies (52) and k ^ 4 , then (57) holds.
[Since R, has I, 1 Fibonacci-number factors, and since

(1 + l ) k = 2 k

and

TVi)h.[

j = (1 - l ) k = 0

we see that the numerator and denominator of Q N , by (55), each has 2


Fibonacci-number factors.

Also, by (4), if a = (1/2)(\j5 + 1)

(1/2)(N/5" - 1) so that a > 1 >. b = l / a


a and b ] ,
(58)

and

k-i

b =

[Carmichael write^ a and -p for my


|
I

a n ( l - b2) ^ a n ( l - b 2n ) =s ^SF^

^ a n ( l + b 2n ) ^ j a n ( l + b 2 )

Therefore, since (1 - b 2 ) / ( l + b2) = 1/V5 and by (55) and (58), Q N ^ a^l/Vs) 2

where, by (36),
f = N - 2N ( 1 ) + SN ( 2 )

-...=N

HH^-KH

234

[Oct

ON THE DIVISIBILITY PROPERTIES

so that

(59)

QN * a ^ V ^ / "

a^^^-^'^^^tVvs)^"

Clearly (qt - l)(q 2 - 1) (q k - 1) exceeds the value when we put q1 = 2 and


qi =2i - 1 (i - 2), namely qk~* (k - 1)1. The function
k

2+

ZX^

i=i

increases more slowly with each q. than does the product, and its value at
x
k
2
the minimal point is 2" + k - k + 1. If k ^ 4, this is seen to be less than
2k~1(k - 1)1. Thus, by (59),
8

(60)

QN ^ | |

-l

|k

I (aV^5) 8

i=i

The function a

/ n has a minimum, for integer values of n when n = 29 and

it exceeds one when n - 4e Thus, by (60),


(61) Q N / q i q 2 o e * q k ^ (a/2)(aV3)(ay5)(a 6 /7)(aY'sy5) 8 = a 2 Vl31250 > 8 ,
and the lemma follows.]
Lemma 23? If N satisfies (52) and k = 3 $ then (57) holds if at least
one q. 11, or if no q. = 2, or if any n. ^ 20
[ As in the proof of Lemma 21, (59) still holds. Now, if we suppose that qt <
q2 < %, we see that qt 2, q2 3, and, by the first supposition of the lemma,
q3 ^ 11. Thus
(62)
(qi - l)(q 2 - l)(q 3 " 1)

:=

(Qt - l)(q 2 - m%

- 2) + (q* - l)(q 2 - 2) + ( q i - 1)

^ 2(q3 - 2) + (q2 - 2) + (qt - 1) * (qt - 1) + (q2 - 1) + (q3 - 1) + 7

1966]

. OF FIBONACCI NUMBERS

.235

and s o ? by (59) and (62), a s b e f o r e ,


(63)

QN/qiq2q3 >

(a/2)(a2/3)(a*o/l 1 ) ^ / 5 * ) =

a 20/l65O

> 9

and (57) follows,


Adopting the second supposition, we have q t ^ 3 q2 > 5 5 and q 3 >

Then (62) Is r e p l a c e d by
(64)

(q t - l)(q 2 - l)(q 3 - 1) > 8(q3 - 2) + 2(q 2 - 2) + q 1 - l 2> (qt -1)+ (q2 - 1 ) +


+ (q3 - 1) + 36 ,

and (63) by
(65)

QN/qiq2(l3 -

(a 2 /3)(aV5)(a 6 /7)(a 3 6 /5 2 ) = a 4 8 /2625 > 10 6

and (57) follows again e


F i n a l l y , if any n. ^ 2 9 0(n) ^ 2(qA - l)(q 2 - 2)(q 3 - 1)0

Thus, as before,

qt ^ 2 5 q2 > 3 , q3 ^ 5 9 and 2 ^ - l)(q 2 - l)(qg - 1) ^ (qt - 1) + (q2 - 1) +


(q3 - 1) + 9; whence
m

%/<li<k<k*

(a/2)(aV3)(aV5)(aV52) = a " / 7 5 0

2.-9

and we get (57).']


L e m m a 24,
33

If N s a t i s f i e s (52).and k = 2 9 then (57) holds If N/qj,q 2 >

o r if at l e a s t one q. > 1 1 .

[ L e t N/q 4 q 2 - r .

Then by (55),

whence, by L e m m a 2,

(68)

Q N > (F

QR = F

r_1

/F^1"1.

/F^Fq^

, and by (8),

T h u s , by (58)

Q N / q i q 2 * { a ^ - ^ " 1 (1 - b ^ V d

+ b ^ ^ / q f t

236

ON THE DIVISIBILITY PROPERTIES

[Oct

First we assume that r ^ 3. Then5 by the kind of argument used, above, if


qt ^ 3 and q2 - 2, and by (68),

(69)

Q N /q x q 2 ^ ( a q ^ / q i ) ( a 3 q 2 " 4 / q 2 ) {a(l - b>)/(l + h*)}^'1


> a 4 (0 o 94) 2 /6 > 1 .

Next, we assume that qt ^ 2, q2 =11, r ^ 1 Then, by (68),

(70)

Q N / q i q 2 ^ ( a ^ ^ ^ / q i X a ^ ' V q z X a a - b 2 )/(l + b 2 ) } ^ " 1 >a9(0 o 72)/22>2.

The results (69) and (70) establish the lemma.]


Lemma 25, If q is prime and q 3, then there is a prime p such
that q = a(p)a
[if q 3, F

2, by Lemma 2, and so F

has a prime factor pe By Lem-

ma 8, a(p) divides q, whence, since q is prime, a(p) = q.]


Lemma 26c If q is prime and X ^ 2, then there is a prime p f 5,
such that q^ = a(p)0
[By Lemma 16 and Theorem 1, if q

= m,

(F

/F

,F

) = 1 if q * 5;

and if 5 ~* = m, ( F 5 m / F m , F m > = 5e If q * 5, by Lemma 17, F q m / F m >


F 4 / F 2 = 3; so that F

must have a prime factor p f 5, prime to F .

If q = 5, since F 2 5 / 5 F 5 = 3001, by (2), Lemma 17 shows that again F


has a prime factor p + 5, prime to F Thus, by Lemma 8, for any q,o?(p)
divides qm = q^ but not m = q " , Therefore q^ = a(p). ]
We now have sufficient information to prove CarmichaePs theorem.
Theorem 60 If N + 1, 2, 6, or 12, then there is a prime p such that
N = ar(p).
Proof,

Let the (unique) prime-power expansion of N be given by (52).

By Lemma 21, Lemmas 22, 23, and 24 show that the theorem holds in the following cases: (i) if k ^ 4, all N; (ii) if k = 3 and either (a) one q. ^ 11,
(b) no q. = 2, or (c) one n. ^ 2; and (iii) if k = 2 and either (a) N/q 1 q 2
- 3 or (b) one qt ^ l l e

In addition, Lemmas 25 and 26 show that the theorem

holds (iv) if k = 1 and N + 28 We see from (2) that, indeed, when N = 1,2,

1966]

OF FIBONACCI NUMBERS

237

6, or 12, there is no prime p such that N = a(p)0

It therefore remains to

show that such a p exists, (v) when k = 3,

no q. 11, one q. = 2, and

no n. 2, and (vi) when k = 2, N 4 6 or 12, N / q ^ = 1 or 2, and no q.


11. We look for primes p which divide F M but no corresponding F
for then N = or(p), as explained earlier,,

,
l(1)

'

Case (v). We have N = 2 - 3 5 = 30, 2 . 3 7 = 42,

and 2 5 7 =

70, We see from (2) that 30 = or(31), 42 = <*(211), and 70 = a(71) = a(911);
so that the theorem holds
Case (vi). We have. N = 2 5 = 10, 22 5 = 20, 2 7 = 14,. 22 7 = 28,
3 5 = 15, 3 7 = 21, and 5 - 7 - 35.
20 = ar(41), 14 = #(29), 28 = ar(281),

We see from (2) that 10 = or<ll),

15 = or(61),

21 = ar(421), and'35 =

a(141961). This completes the theorem,,


Lemma 27. If N = a(p) and N + 5 whence p =(= 5), p Is a

simple

prime factor of FN
[By Lemma 18,

if p Is a mpf of F N , p divides both N and F^0

Thus,

since, by Theorem 1, if p ==
t 5, (p,or(p)) = 1; N must be divisible by por(p),
so that N # a(p)0

The lemma follows.]

By Lemma 27, Theorem 5 Is seen to follow from Theorem 60

We also

see that Theorem 3 and its corollaries follow from Theorems 1, 2, and 6 (with
the exception of the fact that the a (m,p) increase with s).
s
For completeness, we also state the following result.
Lemma 28Q If f1 = 1, f2s f35 ,f = N are all the divisors of N, then

(71)

F N = ]~[Qf
r=i

[If N satisfies (52), its divisors are the (nt + l)(n 2 + I) 8 0% + 1) = m integers
s,k
s
f = q f A q 2 29 q k
where

0 ^ s. ^ n., I = l , 2 > Q ' o , k .

By (55), a particular factor

appear only once in Qf; and this, when

can

238

ON THE DIVISIBILITY PROPERTIES


ti to

g = v

q22

*k

* * ^k

and t. = n. except when i = i l s i 2 s s i t


gq.

or gq. q.

[Oct.

or or gq. q. q.

(when t. < n . ) , only if f = g or


a

It follows by (55) that F

appears

in

n
r=i

to the total power

and 1 if h = 0o This proves that the product is simply F ]


5e PERIODICITY OF RESIDUES
We shall complete this discussion of divisibility properties with a survey
of results pertaining to the characteristic number M(m,.n) defined in Section 1.
Lemmas 13 and 14 show that we may limit the study of the functions
c*(m,n) and M(m,n) to that of ar(p,n) and ju(p,n),

where p is prime.

have established the essential properties of ar(p,n) in Theorem 10

We

Thus, by

(15), the corresponding behaviour of pt(p,n) is known if we know that of /3(p,n)


So far, we have only stated, in Lemma 10, that /3(m) (and, in particular,
/3(p,n)) is always an integer.' The papers of Robinson [ 5 ] , Vinson [ 6 ] , and
Wall [7] have answered almost every question that may be asked about /3(p,n),
and it is their work which will be outlined here c Proofs of all the results quoted
below may be found in Vinson f s paper [ 6 ] , and so will be omitted here,
Theorem 70 If p is an odd prime and n a positive integer, then

4 if pot 2 a(p) = 0 j
1 if pot2c*(p) = 1 (
2 if pot 2 a(p) ^ 2]

1966 ]

OF FIBONACCI NUMBERS

239

but
(73)

(2,1) = 0(2,2) -

1,

and

(3(2,n)

= 2 if n > 3

We note that; with the two exceptions given in (73), 0(p,n) i s independent of n0

Also ?

0(p,n)

always t a k e s one of the t h r e e values

1, 2, o r 4

a remarkably simple result,


T h e o r e m 8e
4,

If m i s a p o s i t i v e i n t e g e r satisfying (27), then (i) j3(m) -

if m ^ 3 and a(m)

i s odd; (ii) j3(m) = 1,

if

pot 2 or(p.) = 1 for e v e r y

p . + 2 (i = 1, 2, , k) and if pot 2 m ^ 2; and (iii) 0(m) = 2 for all other


me
We note that T h e o r e m 8 contains T h e o r e m 7, a s a s p e c i a l c a s e , when
m = p , where p is prime0
T h e o r e m 9a

(The connection i s through L e m m a 13,)

If p i s an odd p r i m e , not equal to 5, and

a positive

i n t e g e r , then
1 if p = 11 or 19 (mod 20)
(74)

j8(p,n) = { 2 if p s

3 or

7 (mod 20)

4 if p = 13 o r 17 (mod 20)
and (of the r e m a i n i n g v a l u e s of p s 1 o r 9 (mod 2O))0(p,n) # 2

if p = 21 or

29 (mod 40) o
T h e s e r e s u l t s a r e connected with the foregoing by way of L e m m a 12.
Vinson points out that the theorem, is " c o m p l e t e " in the s e n s e that e v e r y r e maining possibility o c c u r s ; he l i s t s the e x a m p l e s :
0(521) = 1, 0(41)

= 2, 0(761) = 4, [ p = 1 (mod 40)]

0(809) = 1, 0(409) (75)

0(101) = 1, 0(61)
0(29)

2,. 0(89)

= 4, [ p = 9 (mod 40)]

=4,

= 1, 0(109) -

4,

[ p = 21 (mod 40)]
[p = 2 9

(mod 40)]

REFERENCES
1.

R Da C a r m i c h a e l , "On the N u m e r i c a l F a c t o r s of the A r i t h m e t i c F o r m an


0 n , Ann, Math,', 15 ( 1 9 1 3 - 1 4 ) 3 0 - 7 0 ,

240

ON THE DIVISIBILITY PROPERTIES


OF FIBONACCI NUMBERS

[Oct. 1966 1

20 R D Carmichael, "A Simple Principle of Unification in the Elementary


Theory of Numbers," Arner. Mathc Monthly 36 (1929) 132 - 1 4 3 .
3C J H0 Halton, !TOn a General Fibonacci Identity," Fibonacci Quarterly, 3
(1965) 1: 3 1 - 4 3 .
4. E0 Lucas, "Theorie des Fonctions Numeriques Simplement Periodiques, ,T
Arner, J. Math, 1 (1878) 1 8 4 - 2 4 0 , 2 8 9 - 3 2 1 .
5. D. W0 Robinson, n The Fibonacci Matrix Modulo m, ! f Fibonacci Quarterly,
1 (1963) 2: 2 9 - 3 6 .
60 J0 Vinson, "The Relation of the Period Modulo m to the Rank of Apparition
of m in the Fibonacci Sequence/ 1 Fibonacci Quarterly, 1 (1963) 2: 3 7 - 4 5 .
7. D. De Wall, "Fibonacci Series Modulo m, " Amer, Math. Monthly 67 (1960)
525-532.
* *

All subscription correspondence should be addressed to Brother U. Alfred,


St. Mary ? s College, Calif. Ail checks ($4.00 per year) should be made out to
the Fibonacci Association or the Fibonacci Quarterly.

Manuscripts intended

for publication in the Quarterly should be sent to Verner E. Hoggatt, J r . ,


Mathematics Department, San Jose State College, San Jose, Calif. All manuscripts should be typed, double-spaced. Drawings shouldbe made the same size
as they will appear in the Quarterly, and should be done in India ink on either
vellum or bond paper. Authors should keep a copy of the manuscripts sent to
the editors.
* * *

NOTICE TO ALL SUBSCRIBERS!!!


Please notify the Managing Editor AT ONCE of any address change. The Post
Office Department, rather than forwarding magazines mailed third class, sends
them directly to the dead-letter office.

Unless the addressee specifically r e -

quests the Fibonacci Quarterly to be forwarded at first class rates to the new
address, he will not receive It. (This will usually cost about 30 cents for firstclass postage.) If possible, please notify us AT LEAST THREE WEEKS PRIOR
to publication dates: February 15, April 15, October 15, and December 15.
* * * it it

GENERALIZED FIBONACCI SEQUENCES


ASSOCIATED WITH A GENERALIZED PASCAL TRIANGLE
V . C . Harris and Carolyn C . Sty les, San Diego State Col lege & San Diego Mesa Co liege

1.

INTRODUCTION

In this paper we introduce the numbers


n - ip

Un(P,q,s) =

(1.1)

Lp+sqj

=E

i=o
o = u 0 (p,q,s) == 1

' (

where n, p, q, s are positive integers and [x] is the largest integer in x.


The characteristic equation and a generating function are developed and the
relation to a generalized Pascal's triangle is exhibited in Section 2. An interesting feature is the repetition of each term g times where g = (p, s). Certain sums and some properties relating to congruence are established in
Sections 3 and 4e
The numbers corresponding to the case s = 1 are developed in our p r e vious paper [2] . Thus the Fibonacci numbers are those for p = q = s = 10
The numbers in Dickinson [1] are the special case p = a, q = l

s = c - a.

By multiplying the binomial coefficients

by a11 x % i ( l

before summing, the numbers could be generalized further,

*A more appropriate choice of exponents, suggested by DrP David Zeitlin,


appears in a paper by him which will follow,,
241

242

GENERALIZED FIBONACCI SEQUENCES ASSOCIATED

[Oct

2. THE CHARACTERISTIC EQUATION AND GENERATING FUNCTION


We note that

-flN^WCH*]
iq - 1

or zero, from properties of binomial coefficients.

Hence, if Ef(x) = f(x + 1)

we have (E - 1)1^ is a sum of binomial coefficients with first coefficient involving iq - 1. After repeating q times there results
(E s - l) q u n (p s q,s) = u n _ p (p s q,s) y n - p > 0
or
^

V p + q s = l l(^l]
) VUpn+p+(q-l)s
4 - ( q - l ) s - (\ ?2)/ UVn p % 2 ) s

+ ( 4 ) q + i

'"

+ U

Hence the characteristic equation is


XP(X

(2.2)

_ fi _ i = o

with initial conditions


(2.3)

u0 = u t

p+qs-l

It may be remarked that u ,


=2.
J
p+qs
Suppose the arithmetic triangle to be written but with each row repeated
s times,

Then one sees that u (p,q,s) is the sum of the term in the first col-

umn and n

row (counting the top row as the zero

row) and the terms

obtained by starting from this term and taking steps p, q that i s , p units
up andWhen
q units
to the
(p,s)
= gright.
> l the sequences { u n g } ,

{u n g + 1 }, , K ^ g ^ i

are the same since each sequence is determined by the same recursion formula
and the same initial conditions.

1966]

WITH A GENERALIZED PASCAL TRIANGLE

243

Let f(x) = x P ( x S - l ) q - 1 so that f'(x) = x 13 " 1 (x S - l ) q " j [(p + qs)x S - p ].


The r o o t s of f f (x) = 0 a r e the r o o t s of
x = 0,

xS = 1

xS = E
p + qs

and

None of the r o o t s of ff (x) = 0 i s a r o o t of f(x) = 0 and f(x) h a s no multiple


root.

If the p + sq r o o t s of f(x) a r e x 1 , x 2 , x 3 , , x

ant of the coefficients

c-,. c 9 . , c ,

then the d e t e r m i n -

in

p+qs
p+(is
*V*
n+i
,
.,
c X
= u
^
i i
n
n = 0 , 1 , - , ? + sq - 1
i s different f r o mi=iz e r o . The s y s t e m can be solved by C r a m e r ' s r u l e u s i n g
Vandermondians.

It r e s u l t s that c. = (x. - 1 ) / [(x, - l ) { [ p + s q ] x . - p } ]

hence

2 4

( - >

/ s
., n+i
p+sq
(x. - l v) x
L
= E
"
* n=
i=i (Xi - l ) [ ( p + sq)xf - p ]

0,1,2,

To obtain a g e n e r a t i n g function, w r i t e

S = ] u i x l
Then by multiplying S by each of
<-l)(;)xS, ( - l ) * ( j y

,-.(-Dq(j)x*

and

. P^qs
-: x

and adding, one finds

f (i-x^- x p

+sq

q-i s - i

-/ \

] S = i:zi:(-i)Y?)x ks+i
K

k=o i=o j=o

V/

Note we have u s e d u 0 = ut = - = Up+ S q> 1 = 1 and (201)

Hence

and

244

GENERALIZED FIBONACCI SEQUENCES ASSOCIATED

y
_^ /..^

\
n=o

JLm~** *

q
\ ]

[Oct.

^^
I X

\_71'i
u xii _=_ k- < i- J ri
"(3 " ' j / j ' i T / ^

The numerator is equal to


s-i

/ 0-1

s 1

"

S^g'-^v)^}-<-*>-'
= (1 - x s ) q / ( l - x)

Hence

f\

(2.5)

.fe

= (1 - xS) q /(l - x)

(1 - x) q - xP + s q

As an example, for p = 2, q = 2, s = 3, this gives

7 %
^e

(2,2,3) x11 = ^ - +

X + X2

- ^ " x4 "
1 - 2x3 + x s - x8

x5

This gives the sequence


{u n (2,2,3)} = 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 2 , 2 , 2 , 4 , 4 , 4,7, 7, 8,12,12,16,21,21,31, 37,.'

3. SUMS

(3.D

+
XX.-I!
(VfV
)r
---
v
i=0
i=o
u

n+p+s(q-i)-i

where ^y is Kronecker T s 6.

'' *+

n +p+s(q-i)-s+i

J(

S<5l

1966]

WITH A GENERALIZED PASCAL TRIANGLE

245

This is seen to be true for q = 1 and all n by summing u0 = u


V

Ul =

Vsq-i

Vs+i - V i ; " * ;
= l j

thiS

Vs+n " Vtf

+s

Since

u = U l = . . . =

giv6S

XX

n+p+s

n+p

E i - (p > - 5 > i + p

i=o

i=o

i=o

which is the result. Also this is true for n = 0 and all q. We have to show
q_1

i/
i\
o = (-Dl(q ~i M fVs(q-i)
i=o
A
/

But u

'

Vs(q-i)-(s-i)J-

= 2 so that by separating the term corresponding to i = 0 we get

l + s ( l - l)^-1 = 1 = u0.
It remains to show the result in general.

Assume (3.1) to be true for

q > 2 and n = k ; then


k+i

EUi

k
=

Vi + E ui

i=o

E ( " 1)l (?)Vpf(q-l)B+i

i=o

i=o

i/ - A

*"'

2J(-1) I
i=o

i lrVp+s(q-i)

+ U

k+l+p+s (q~i)

k+p+s(q-i)+i

using

k+p+s(q-i)-i

By combining terms
U

>

the result follows and the theorem is proved,

'

'"

k+p+s(q-i)-s+iJ

246

GENEEALIZED FIBONACCI SEQUENCES ASSOCIATED


q n+p+s(q-k)
tE ' '[E '" ( - l ) " + I , + B < C ' - k ) + k - i n 1
k=o

eve.

i=n+i
q n + P t s(q-k)

1=0

[Oct.

(-l)

P+Sq

k=Q

I=m-i

n+^s(q_k)+k-i

(-1)
0

/ 1 \

P=q=l (mod 2)'

+ \ ( - l ) n + 1 2 q - 1 p=q=o (mod 2) sodd


! (-l)n

(3.2)

Proof: Solve (2.1) for u

2 q " 1 pq

(mod 2)J

and write (-1) ~ u. for i = n, n - l , e " , 0 .

The

sum of the (k + l)st column formed by the expansions is

n+p+s(q-k)

1=0

'

1=0

- ( - l ^ Y ^ t - 1 + -.. + (_i)lH-B(q-k)-l]

since the terms added to obtain the sum on the right have u. = 1 in each case.
Hence this gives
n

n+p+s(q-k)

s>'"(;)v - E "*"(;),
i=o

'

i=o

+ (-l)n+kl

where e = 0, p + s(q-k) = 0 (mod 2) and = 1 otherwise.


Summing for k = 0 , l , * # # ' , q gives

I'

1966]

WITH A GENERALIZED PASCAL TRIANGLE

q n+p+s(q-k)

Y)-!)"- u. = y

As**

JLmmJ

smaJ

i=o

k=o

247

(.D^stq-^k-i / \
\ If /
x 7

i=n+i

+y

V(-ir pf8(q - k)+k - i f?)u. +

k=o i~o

) k=o
S ( " 1)R+k \k/ fk)

J P + S(q

"" k)* 0 ^ m o d 2 >

, p + s(q - k) = O(mod 2)

But
q

y^^^jn+i^stq-kj+k-i/
J 4 S X# SJHBSSW

Ju.

0 ,

s s 0 (mod 2)
n
fB
q2qJ](-i)n-1u1,8Ei(iiiod2)
(_i)p
i=o

and

V ^ y -,,n+k/q \

k)

~ l ) n + 1 2 q " ' 1 , s odd p = q = 0 (mod 2)


-l)n

2 q " 1 , s odd

0
Combining t h e s e r e s u l t s gives the t h e o r e m .
It may b e r e m a r k e d that

p*q

, otherwise

(mod 2)

248

GENERALIZED FIBONACCI SEQUENCES


[Oct. 1966]
ASSOCIATED WITH A GENERALIZED PASCAL TRIANGLE
n
n
/ , u 2i and ; P u 2 i + 1
i=o
i=o

can be obtained from (3.1) and (3.2).


4. DIVISIBILITY PROPERTIES
Using methods similar to those of our previous paper, one can show the
following: Any p + sq consecutive terms are relatively prime 0

The least

nonnegative residues modulo any positive integer m of u (p,q,s) are periodic


with a period P not exceeding m p + s ( ^ There is no preperiod and each period
begins with p + sq terms all unity.

Any prime divides infinitely many

u n (p,q,s) since

P~p

X / ^ M 1 ) UP+s(q-i)
i=o

'

(m dm

>-

REFERENCES
1. David Dickinson, "On Sums Involving Binomial Coefficients," American
Mathematical Monthly, Vol. 57, 1950, pp 8 2 - 8 6 .
2. V. C Harris and Carolyn C. Styles, "A Generalization of Fibonacci Numbers," 2 ] j e j l b a a a ^ ^
Vol. 2, 1964, pp 277289.

*****
REQUEST
The Fibonacci Bibliographical Research Center desires that any reader
finding a Fibonacci reference send a card giving the reference and a
description of the contents.

Please forward all such information to:

Fibonacci Bibliographical Research Center,


Mathematics Department
San Jose State College
San Jose, California

brief

CORRIGENDUM FOR "SOME CONfERGENT RECURSIVE SEQUENCES,


HOMEOMORPHIC IDENTITIES, AND INDUCTIVELY DEFINED COMPLEMENTARY SEQUENCES"
John C* Hollcfdcty^ Institute for Defense Analysis^ Washington, D . C .

On the above-entitled paper, appearing in the February 1966 volume of


the Fibonacci Quarterly, please note the following changes:
Page 130 The last two lines of the Corollary should read:
0 . 0 and only one homeomorphism g such that g > 1 and
(2.9)

h + h" 1 = g + g" 1 .

Page 14. Equation (2a27) should read:


(2027)

(hUh" 1 )^) = t for all (hUh~ 1 )" 1 (x) < t < x0

Equation (2e3Q) should read:


(2o30)

h n + 1 = P - IT 1

n > 1.

Equation (2.31).should read:

(2.31)

hjI -

II

Page 15. Equation (2e36) should read

(2.36)

hj(^H) I

Page 16. Equations (2e39) and (2.40) should read as follows:

,2.S9,

v % I ^, B
249

= (i#ii)

250

CORRIGENDUM

(2.40)

[Oct 1966]

a + \a2 - 4 J _
2
/

Page 21. After proof for the Corollary, add a Reference [5].
Page 23. The first line of the Corollary should read:
Corollary; Let P(n) t 2n for some integer n > 0. Then
Page 24. Change the last line of Theorem 16 to reads
{ x } be inductively defined by
Equation (3.26) should read(3,26)

x0 < a( X-1 ) - x_1

Page 26a Equation (4.13) should read:


(4.13)
F(x) = (a - ^a2 - l)x + 0 - P'JaT^i/{a
Page 27. Equation (4.19) should read:
(419)

~ 1)

(x - p) {(a - l)x + pa + p} > a2e/(a + 1)

Equation (4,24) should begin with the line


(424)

hf *(x) = xF(l)

0 < x < 1

Page 28. The first line on the page should read:


If p > 1* , then F(l) > 0 implies that -e < (p - I) 2 . It may
be
Page 29. Equation (5.3) should read:
(5.3)

hh(t) < gh(t)

Page 33. Change the first line of Theorem 22 to read:


Let /J < 1 and P +/xl > I. Let gt be any
Page 34. The last three lines before the Corollary should read:
for h 1 has been proven. To prove convergence for g l s insert \L into the proper
positions of (1.39) and (1.40), and continue the argument of the paragraph containing (1.39) and (1.40). Uniqueness of h is obtained from Theorem 21.
Page 36. Add References below.
11. J. Lambek and L. Moser,, Inverse and complementary sequences of natural numbers, Amer. Math. Monthly, 61 (1954), 454-458.
12. H. W. Gould, ff Non-Fibonacci Numbers, ff Fibonacci Quarterly, Vol. 3,
No. 3, October 1965, pp. 1 7 7 - 1 8 3 .
* * it is

IDfllCED PIOILEiS AND SOLUTIONS


Edited by V e E e Hoggatt, Jr e / San Jose State College, San Jose, Calif.

Send all communications concerning Advanced Problems and Solutions to


Verner E0 Hoggatt, Jr 0 , Mathematics Department, San Jose State College, San
Jose, California,,

This department especially welcomes problems believed to

be new or extending old results.

Proposers should submit solutions or other

information that will assist the editor 0 To facilitate their consideration, solutions should be submitted on separate signed sheets within three months after
publication of the problems,,
H - 8 9 . Proposed by Maxey Brooke, Sweeny, Texas.

Fibonacci started out with a pair of rabbits, a male and a female 0

female will begin bearing after two months and will bear monthly thereafter,,
The first litter a female bears is twin males, thereafter she alternately bears
female and male.
Find a recurrence relation for the number of males and females born at
the end of the n
H-90

month and the total rabbit population at that time.

Proposed by V. E. Hoggatf, Jr., San Jose State College, San Jose, Calif.

Let the total population after n time periods be the sequence |F^} __
determine the common birth sequence for every female rabbit and tie It in with
the value of the Fibonacci polynomials at x = 20
x

fn+2( )

xfn+i( )

H - 9 1 . Proposed

Let m =

+ f

n( >'

(f0(x) = 0, f^x) = 1 and

* 0).

by Douglas Lind, University of Virginia, Charlottesville,

, then show

m=i
F

// Fx

( 1 JnL
<k-1-2j + e n >
kn n
n = E
Z ~ ^ - >' TE-1-2J

where
251

Va.

[Oct

ADVANCED PROBLEMS

252

| ( - l ) m n if k is odd
f

if k is even

and [x] is the greatest integer not exceeding x0


H-92

Proposed by Brother U. Alfred, St Mary's

College, Calif.

Prove or Disprove: A part from F 1? F 2 , F 3 , F4> no Fibonacci number,


F. (i > 0) is a divisor of a Lucas Number
H-93

Proposed by Douglas Lind, University of Virginia, Charlottesville,

Va.

Show that
n-i

n . g (.*.. ^ )

F =

k=i
n-2

,-E( + ^ ^

L. =

k=i

where n is the greatest integer contained in n/2 c


SOLUTIONS
H - 5 0 Proposed by Ralph Greenberg, Philadelphia, Pa. and H. Winthrop,

University

of So.

Florida, Tampa, Fla.

Show

Iln, = F.2n

n1+n2+n3+ +npn
where the sum is taken over all partitions of n into positive integers and the
order of distinct summands is considered,,

1966 ]

AND SOLUTIONS

253

A paper by D A Lindand V0 E0 Hoggatt, Jr 0 , "Composition Formulas Derived


from Birth. Sequences, !f will appear soon in the Fibonacci Quarterly, and will
discuss this among many other examples,,
H-22

Proposed by V. E. Hoggatt, Jr.r San Jose State College, San Jose, Calif.

If

oo

P(x) =TT(1

+ xFi

i=i

y^R(n)xn
n=o

then show
(i)
(ii)
H-53

R(F 2 n - 1) = n
R(N) > n if N > F 2 n - 1

Proposed by V. f . Hoggatt, Jr., San Jose State College, San Jose, Calif.
and S.L. Basin, Sylvania Electronics Systems, Mt. View, Calif.

The Lucas sequence

L t = 1, L2 = 3; L Q+2 = L n + 1 + L n for n

1 is

incomplete (see V E0 Hoggatt, JrB and C0 King, Problem E-1424 American


Monthly, Vol. 67, No0 6, June-July I960, p0 593, since every integer n i s not
the sum of distinct Lucas numbers 0
cannot be so represented.

OBSERVE THAT 2, 6, 9, 13, 17,

Let M(n) be the number of positive integers less

than n which cannot be so represented.


M(Ln) =

Show
Fn.t

Find, if possible, a closed form solution for M(n)

A paper by David Klarner to appear soon in the Fibonacci Quarterly completely answers these questionsc
H-55

Proposed by Raymond Whitney,

Lock Haven State College, Lock Haven, Pa.

[Oct.

ADVANCED PROBLEMS

254

Let F(n) and L(n) denote the n

Fibonacci and n

Lucas numbers,

respectively,
Given U(n) = F(F(n)), V(n) = F(L(n)), W(n) = L(L(n)) and X(n) = L(F ),
find recurrence relations for the sequences U(n), V(n), W(n), and X(n)
A paper by student Gary Ford to appear soon in the Fibonacci Quarterly
offers several answers to this problem,,

Also a paper by R. Whitney deals with

this and will appear shortly,


H-52

Proposed

by Brother

U. Alfred, St Mary's

Co!lege / Calif.

Prove that the value of the determinant

n+2

n+4

i2
n+2

u2
n+4

n+6

n+4

n+6

n+8

is 18(-1)"n+i
Solution by y, E. Hoggatt, Jr., San Jose State College, San Jose, Calif.

Since F 2 = (L

- 2)/5, the auxiliary polynomial satisfied by F 2

is the

product of the auxiliary polynomials for L : (x - 7x + 1) and for C = -2 :


(x - 1) or
(x2 - 7x + l)(x - 1) = x3 - 8x2 + 8x - 1
Therefore the recurrence relation for F 2

is

2n

u2 ,. = 8u 2 , J - 8u 2 , n + u2 .
n+6
n+4
n+2
n
Thus D

= (-l)D

by using the above recurrence relation after multiplying

the first column of D by - 1 . ' The value .of D0 is -18, therefore D = 18(-1)
TWO BEAUTIES
H-47

Proposed

by L. Carlitz,

Duke University,

Durham,

N.C.

1966]

AND SOLUTIONS

255

Show that

V (n

S\

* I

n=o

n
nX

M*>
(1 - x - x2)k

where

- ! > ' Ck) v

k'

r=o

H-51

Proposed fey V. . Hoggatt, Jr., San Jose Sfofe College, Son Jose; Colli.
one/ L. Carlitz,

Duke University,

Durham,

N. C.

Show that If

xt
=
1 - (2 - x)t + (1 - x - x2)t2

(I)

Z Q k^ t k
k=1

and
OO

\w

(ii)

(1 - x - x 2 ) k

n=o
that

*k = E ( " 1 > r ' 1 (r) F ^ P

= Q

k< x >

r=o
Solutions by Kathleen Weland, Gary Ford, and Douglas Lind,
Program, University

of Santa Clara, Santa Clara,

Calif.

Undergraduate

Research

256

[Oct

ADVANCED PROBLEMS
It is familiar that

J?L

/n + k - 1\

Let W n obey W n+2 = pW n + 1 - qW , p 2 - 4q 4 0, and let a b both satisfy


x2 - px + q = 0, so that a + b = p., ab = q. Then W = Aa11 + Bb11 for some
constants A and B and all n. It follows that
*

/n + k- 1\
v

A(l - ax)

+ B(l - bx)

A(l - bx) k + B(l - ax) k


(1 - px + qx2)

A
-

'-

i)j

X/- (?)

by+B

3=0

l)j

Z)"
fi)
EL

aJ

(1 - px + qx 2 ) 1

2>J0)
4=o

- 4zL

(Aa

+ Bb J )(abx) 3

(1 - px + qx2)"

(1 - px + qx 2 ) k

%(x)
(1 - px + qx 2 ) k

Toget(ii)ofH-51weput p = l , q = - 1 , W = F , andrecalling F_ = (-1)" F


we have
00

n=o

l"k
j=o

(1 - x - x 2 r

1966"

AND SOLUTIONS

To obtain H-47,.we s e t p = 1, q = - 1 , W

257

= L , and r e m e m b e r i n g that L_

= v(-1) L we find
'
n

'n + k - 1
T

L x
n

n=o

y%l)J

(1 - x - x 2 )

i=o

We now g e n e r a l i z e (i) of H-51 Q

R (x) =

171
LJC
1/
J

Since

l > q ) j ( J) w-fj

= A(l - b x ) k + B ( l - a x ) k

we have

y^Rk(x)tk = A ^
k=o

pi - bx)t]k + B V [ ( 1

k=o

- ax)tf

ko.

1 - (1 - bx)t

B
1 - (1 - ax)t

A + B - (A + B)t + (Aa + Bb)x1'


1 - (2 - px)t + (1 - px + qx 2 )t 2
Now W0 = A + Bj

R (x)tk =

(*) X k
k=o

Putting p = 1,

Wj = Aa + Bb,

so we may w r i t e

W0 + (xWj - W 0 )t
1 - (2 - px) t + (1 - px + qx 2 )t 2

q = - 1 , W n = F ^ m a k e s R k (x) - Q t (x) of P r o b l e m H - 5 1 ,
n
n
and (*) b e c o m e s

258

ADVANCED PROBLEMS AND SOLUTIONS

[Oct. 1966 ]

xt
1 - (2 - x)t + (1 - x - x2)t2

E Qktk
k=o

the required r e s u l t
Also solved hy the proposers.

LATE PROBLEM ADDITIONS


A SIMPLE PROOF, PLEASE!
H-91f Submitted by RobertW. Floyd, Carnegie Institute of Technology, and
Donald E. Knuth, California Institute of Technology,
Let a be any irrational number, and let the notation {xj stand for the
fractional part of x. Suppose a man has accurately marked off the points 1,
0, \a>\, \2a},9

j(n - l)a\

on a line, n > 1. These n + 1 points divide the

line segment between 0 and 1 into n disjoint Intervals.


man adds the next point \na\,

Show that when the

it falls in the largest of these n intervals; if

there are several intervals which have the maximum length, the point \na>\
falls in one of these maximal intervals,,
ratio

?f

Furthermore, If a is the

!?

golden

</>"" = y ( v ~ - ' l ) = 0.618 , then the point {nor} always divides the

corresponding interval Into two intervals whose lengths are In the golden ratio.
A number a has the property that {no?} always divides Its interval Into two
parts, such that the ratio of longer to shorter is less than 2, if and only if |or|
= <f>~ or 0 " s

(Notes The fact that the fractional parts |nor| are asymptotic-

ally equidistributed In (0,1) is well known; this problem shows the mechanism
behind that theorem, since {n#| always chooses the largest remaining open
Furthermore, the sequence | n ^ | is the "most equidistributed T? of all

place.

these sequences.)
H-95 Proposed by J. A. H. Hunter, Toronto, Canada.
Show
F3

n+k

Lk =

k[

F F

k 3n

^M]

* * * *

'

A SINGULAR FIBONACCI MATRIX AND ITS RELATED LAMBDA FUNCTION


Curtis M c K n i g h t (student) & Dean Priest, Harding C o l l e g e , Searcy, Arkansas

After a very brief introduction to some of the extremely basic properties


of Fibonacci numbers, a student of mine inductively produced the following
identities concerning determinants of Fibonacci matrices:

(1)

n+i
^n+s+3

IF- n+s+2
n

n+m+2

(2)

|
(3)

n+i
= <-l)" F S+2

n+m+i

(- 1 )

n+2m+3

n
F

n+m+s+2

n+m+i |

n+i

m+i F m+2

n+i.
= (-1)'
' m+i * m+s+2

n+2m +S+3!

Each row of the determinant is regarded as a pair of numbers, the subscript


s refers to the number of terms in the Fibonacci sequence skipped between
successive pairs, and the subscript m refers to the number of terms skipped
between the two numbers of a pair,
It is simple exercise to establish the validity of (1), (2), and (3) using
F

m+n

= F

n-iFm

+ F

n F m+i- However, close inspection will show that (1), (2),

and (3) are only special cases and/or variations of


(4)

F F
P

p-k

p-k F q+k

F F

k q+k-p

where k = m + 1 and q - p = s + 1.
This comparison is made easier when (4) is written as

(4')

F ,
p-k
q

F
p
q+k

p-k+i
<-

F F

k q+k-p

thus suggesting a form for a related 3 X 3 matrix


259

260

[Oct.

A SINGULAR FIBONACCI MATRIX AND


F

P ~

J-k

j+k"

Fm-k

Fm

m+k

Fn+k

n-k

J
n

A singular property of the Det(P) presents itself.


Theorem: Det(P) = 0, k, j j n ^ n are integers.
Proof: There is no loss in generality to assume j > m > n and it is simply
convenient to assume k > 0. By applying (4) it is apparent that the columns
of P are linearly dependent. We note by inspection that F ^ (column 3) - F 2 k
k+i
(column 2) = (-1)
F^ (column 1). Thus, the determinant is clearly zero (0).
Q. E. D.
Since the Det(P) = 0, a previous article of this Quarterly [3 j suggests
it would be interesting to consider the Det(P + a) where P + a means a matrix
P with a added to each element of P.

The generality of j , m , n ,

and k

would almost prohibit the techniques used by Whitney [ 3 ] . Hence procedures


discussed by Bicknell in [l] and by Bicknell and Hoggatt in a previous article
of this Quarterly [2] are employed. Using the formula [2]
Det(P + a) = Det(P) + a X(P) ,
where X(P) is the change in the value of the determinant of P, when the number 1 is added to each element of P , we have
Det(P + a) = a \(P) ,
since Det(P) = 0. NowA(P) and the corresponding Det(P + a) are interests
ing in any one of the following forms. They are also derived with the aid of (4T).
1

(a)

MP) =

m-k "Fj-k

or

n - k '" J-k

Fm-Fj
F

n- j

m+k

j+k

-Vk

n+k

1966]
(b)

ITS RELATED LAMBDA FUNCTION


X(P)

[F2k - F k - fi)kF k j| (-1) nx-k^

261

+ (-1) j-fcF

. - (~1)J
n-j v '

F
m

-]

Therefore,

(c)

Det(P + a) =

FTm - k

F,
j-k

n-k ~

j-k

F
F

- F- F-

*m+k " *j+k


F
n+k " F j+k

or
m T
(d) Det(P+a) = [ F 2 k - F k - (-1) F k ][ ( - l )m-k
" X _ m + (-1)0-* F

i-K r
]a.
n-j. - (-1);
m-j

The first factors of (b) and (d) have a straightforward simplification if it is


known in advance whether or not k is even or odd.

The various forms of

X(P) and Det(P + a) become much more intriguing once the interesting patterns in the subscripts and exponents and their relationship to P are observed.
These patterns could easily serve as mnemonic devices.

REFERENCES
1 Marjorie Bicknell, n The Lambda Number of a Matrix; The Sum of Its n2
Cofactors," Amer. Math. Monthly, 72 (1965), pp 260264.
2. Marjorie Bicknell and V. E. Hoggat^ Jr. , "Fibonacci Matrices and Lambda
Functions," Fibonacci Quarterly, Vol. 1, No. 2, April 1964, pp 4750.
3.

Problem B-24 9 Fibonacci Quarterly, Vol. 2, No. 2, April, 1964.

REQUEST
The Fibonacci Bibliographical Research Center desires that any reader
finding a Fibonacci reference send a card giving the reference and a
description of the contents. Please forward all such information to:

Fibonacci Bibliographical Research Center.

Mathematics Department,
San Jose State College
San Jose, California.

* * *

brief

EXPLORING SPECIAL FIBONACCI RELATIONS


Brother U. Alfred, St. Mary's College, Calif.

In a previous exploration section, readers were introduced to the problem of generalized Fibonacci-Lucas relations. We denoted the terms of the
generalized Fibonacci sequence as f
Lucas sequence as g

and those of the associated generalized

where
=

Sn

n-i

n+i

Recall also that the sequence (2, 9) means the Fibonacci sequence with ft = 2
and f2 = 9. The following represent some curious results obtained in trying
to express f g

as a linear combination of f f s and g T s.

Sequence

Formula for

fngn

(2>9)

nn

(3 >7)

nSn

S2n+3 ~ f 2n-i

(3.10)

f n g n = g 2n+3 + f2n +

(3,H)

fn%

(4,13)

f n g n = g 2n + 4 + *2n-4

(5.11)

fngn = S2n+3

2n+3

+ f

(5,13)

f n g n = g 2 n + 4 + f 2 n-i

+ f

M,W>

n&ti

^2n+4

'(6,17)

2n+5 "" f2n+3

(7,17)

f n g n = g 2 n + 4 + g 2n+2 + f2n-4

nSn

2n+5 ~ S2n-2

gm_t

2n+4 ~ S2n - f2n-3

+ f

2n-2
2n-4

S2n-i
+ f

2n-4

Now, of course, it must be recognized that these linear expressions could be


represented in an infinity of different ways. However, it does not seem that
they are all one and the same relation.

If not, then we have specific relations

that characterize the individual sequences. So the following questions are raised:
262

[Oct 1966]

EXPLORING SPECIAL FIBONACCI RELATIONS

263

(1) Can the above formulas for f n g n be unified into one formula?
(2) If not, can other instances be found of this type of phenomenon?
(3) When is it that we have particular formulas for each Fibonacci sequence
rather than a general formula for all sequences?

All subscription correspondence should be addressed to Brother U. Alfred,


St. Mary f s College, Calif, All checks ($4.00 per year) should be made out to
the Fibonacci Association or the Fibonacci Quarterly.

Manuscripts Intended

for publication in the Quarterly should be sent to Verner E. Hoggatt, Jr. ,


Mathematics Department, San Jose State College, San Jose, Calif. AH manuscripts should be typed, double-spaced. Drawings should be made the same size
as they will appear in the Quarterly, and should be done in India ink on either
vellum or bond paper. Authors should keep a copy of the manuscripts sent to
the editors.
* * * *

NOTICE TO ALL SUBSCRIBERS!! I


Please notify the Managing Editor AT ONCE of any address change. The Post
Office Department, rather than forwarding magazines mailed third class, sends
them directly to the dead-letter office.

Unless the addressee specifically r e -

quests the Fibonacci Quarterly to be forwarded at first class rates to the new
address, he will not receive it. (This will usually cost about 30 cents for firstclass postage.) If possible, please notify us AT LEAST THREE WEEKS PRIOR
to publication dates: February 15, April 15, October 15, and December 15.
* * * -k *

The Fibonacci Association Invites Educational Institutions to apply for Academic


Membership in the Association. The minimum subscription fee is $25 annually.
(Academic Members will receive two copies of each issue and will have their
.names listed In the Journal.
*****

ALGEBRA THROUGH PROBLEM SOLVING


BOOK

REVIEW

Brother U. Alfred, St. Mary's College, Calif.

A paperback entitled Algebra Through Problem Solving, written by Abraham


P c Hillman, University of New Mexico, and Gerald L. Alexanderson, University of Santa Clara (1966, Allyn and Bacon) has appeared as one of a series
entitled Topics in Contemporary Mathematics,,

The following sequence of

topics is covered 0
1. The Pascal Triangle
2. The Fibonacci and Lucas Numbers
3. Factorials
4. Arithmetic and Geometric Progressions
5. Mathematical Induction
6 The Binomial Theorem
7. Combinations and Permutations
8. Polynomial Equations
9. Determinants
10o Inequalities,,
The manner of treatment is to give a presentation of the basic ideas in a
succinct and cogent fashion and then allow the student to become familiar with
these ideas by solving numerous and varied problems. This approach, it seems
to the present reviewer, is highly commendable,,
Of particular interest is the role given the Fibonacci and Lucas numbers.
Note that these sequences are introduced early in the bookG In Chapter 2, they
are used in the main to provide the student with an opportunity to make mathematical conjectures on new and unfamiliar material.

But the authors do not

stop there. As might be expected, they play a key role in the chapter on mathematical induction and are brought in occasionally in other portions of the book
as well.
An unofficial and rapid count of these loci apart from Chapter 2 is given
below.
In the chapter on mathematical induction, p. 35, No. 4; p. 37, No. 17,
p. 38, Nos. 19-25 inclusive; in the chapter on the binomial theorem, Nos. 3336, pp. 49-50; in the chapter on determinants, No. 17, pp. 94-95; in the chapter
on inequalities, No. 5, p. 107, No. 28, p. 117.
(Text continued on p. 269.)
264

OPTIMALLY PROOF FOR THE SYMMETRIC FIBONACCI SEARCH TECHNIQUE


Mordecai A v r i e l and Douglass J . W i l d e , Stanford University, Stanford, C a l i f ,

An Important problem in engineering, economics, and statistics is to find


the maximum of a function.

When the function has only one stationary point,

the maximum, and when it depends on a single variable in a finite interval, the
most efficient way to find the maximum is based on the Fibonacci numbers. The
procedure, now known widely as "Fibonacci search," was discovered and shown
optimal in a minimax sense by Kiefer [1 ].
demonstration.

Se Johnson [2 ] gave a different

Oliver and Wilde [3,4] extended the procedure to the case

where, in order to distinguish between adjacent measurements, a non-negligible


distance must be preserved between them. Although this modification, called
the symmetric Fibonacci technique is described informally in [4],

where

numerous extensions are also discussed the present paper gives the first complete, precise description, and formal optimality proof.
Let y(x) be a unimodal function, i. e8 , one with a unique relative maximum value obtained at x = x* on the closed interval
(1)

y(x*) =

[a,b]; thus

max y(x)
a<x<b

and a < xt < x2 < x* implies y(x1) < y(x 2 );x*<

xt < x2 ^ b implies

y(xt)

> y(x2)e As a practical consideration, assume that y(x) canbe measured only
with finite accuracy, but there is a 6 > 0 such that if \xt - x 2 | > 6 the measurements of y(xt)

and y(x2) will be equal only if xt and x2 lie on opposite

sides of x*s
A search strategy S(n, 6) on [a,b] is a plan for evaluating the function
at n distinct points

XUK^9^

,x n , where the location of x^ + 1 depends on

y(xj) for j ^ k and where


(2)

|xj - x k | > 5

j * K

1 < j,

k < n

The plan terminates after successive reduction of a starting interval on which


the function is defined to a final interval of a required length, containing x*e
265

OPTIMALITY PROOF FOR THE SYMMETRIC

266

[Oct.

Suppose k function evaluations have been performed, and let m k be such that
(3)

y(m k ) = max{y(x 1 ),--,y(xk)}

Let
(4)

L k = {xj:xj < mk, j < k } u { a }

(5)

Rk = {xj:xj > m k , j < k}U{b}

and
(6)

k = Sup L k

(7)

r k = Inf R k
Since y is unimodal, it is clear that x* must be in the interval [ k , r k ] .
The purpose of this paper is to derive a search strategy S op (n, <5) that

is optimal in the sense that it maximizes the length of the starting interval for
whatever final interval is given.

From (2) - (7) it follows that the final inter-

val for any n > 1 is


(8)

r n -.* n > 26

By choosing the required length (>2 6) of the final interval we rescalethe s y s tem as follows:
(9)

r n - 7 n = \ ( r n - j e n ) = 1,

(10)

6 =

(11)

x = X(x - a)

\ >0

Thus the new strategy S(n,6*) is defined on [ 0 , D n ] , where D n = X(b - a) 9


and it yields a final interval of unit length.

19

^6J

FIBONACCI SEARCH TECHNIQUE

267

The symmetric Fibonacci search is a search strategy f o r a starting


interval [0,5"^] where the right end point D^ is given by
(12)

DF = F n + 1 - Fn^69

n = 1,2,-..

where
(13)

Fn+2 = Fn+1 + Fn

(F 0 = 0, Fj = 1)

are the Fibonacci numbers.

The strategy is defined by the rules

(14)

x 4 F = Fn - F n _ 2 6

and
<15>

J+i = ' k

+ 7

k -

Lemma: The symmetric Fibonacci search defined above is an S(n,6)


search strategy for [ 0 , D n ] with r

= 1.

Proof; By induction on n The lemma i s trivially true for n = 1, since by


(3) - (7) It = 0 and rA = 1.
_

For n = 2,
_

5 f = 2 - <5 and xt
-p

= 1,imply-

ing that Ht = 0, rt = 2 - 6 and consequently x2 = 1 - 6 by (15). Thus


|x t - x2 I = 6 and for any unimodal function either m2 = xt or m2 = x2 ,
and in both cases r 2 - 12 = 1.
Now assume the lemma true for n = N ^2) a Then D N + 1 = F N + 2 - FN<5
-p

TT

and x t = F N + j - FNmml6 = D N Consequently, j ^ = 0, r t = F N + 2 - FN<5 and


x2F= F N - F N _ 2 6 = H j _ 1 . Also | x f - x f |. =. D^ - D ^
= xf

=> 1 -g" =>g\ If

m2

then T2 = 0 and r 2 = x^ = D^ s Thus 0 < x* ^ D^ with one evaluation

of the function at D^

and G

the induction hypothesis the lemma is true

for this case. If m2 = x ^ , then i2 = x2 = D j ^ and r 2 = D N + 1 . Define


the new variable x F = D J_, 1 - xF so that 0 ^ x F < D with one function
__F
N+l
N
evaluation at D^_x- Thus by induction the lemma is true for this case, too;
Note that if y(x^ + 1 ) = y(i%) and say n% > xj[ +1 , x* is known to lie
in the interval [ x ^ + l s m k ] . As iBf^+i = n% or mfc+1 = xE+, , the Fibonacci

[Oct.

OPTIMALITY PROOF FOR, THE SYMMETRIC

268

search plan suggests reducing the interval [ % J ^ ] to either


[xf + 1 , r ^ ] .

[^,1*%] or

An improved choice could be made in this case by taking the next

interval as [ ^k+l'^k+ll ~ [^l+lk] which can be the starting interval for a


new S(n - k - 2,6) Fibonacci plan.

This fortuitous situation is deliberately

excluded from the theorem following in order to simplify the proof.


Theorem: The symmetric Fibonacci search plan is the unique S (n,6)
strategy among all S(n,6) strategies, provided y(x^ + ^) ==
t y(m^) for every
k = 1,2, , n - 1
Proof: by induction on n. For n = 1 the proof is obvious because all ~Dt =
-F
_
_
Dj = 1. For n = 2 we can assume without loss of generality that x t > x2
Then any S(n,6 ) must satisfy Xj = 1, D2 - Xj = 1 - 6, yielding D2 < 2 - "6 =
Tp

D2 o Assume now that the theorem is true for n = N ^ 2, then for an S(N +
1,6) strategy let Xi > x2 and m2 = x 2 . Hence T2 = 0, r 2 = x1? and we have
an S(N,6) strategy on [O.Xj], By induction
(16)
If m 2

xt * D ^
=

x1? then ^2 = ^ 2 , ^2 ~ D N+1 > and by induction therefore

(17)

B N + 1 - x2 ^

DN

However,
(18)

x2 < D * ^

also by induction, since if x* < x2 then x* has to be located by an S(N - 1,6 )


strategy on [0,x 2 ].

Thus addition of (17) and (18) yields

<19>

N+1
jp

Tp

S 5

N + l-l

Tp

But D N + D N - 1 = D N + 1 by (12), and this shows that the symmetric Fibonacci


search is indeed optimal.
Moreover, we observe that
(20)

D N + 1 - xt ^

196

6]

FIBONACCI SEARCH TECHNIQUE

269

by an argument similar to that used in (18).' Relations (16) - (20) show that
the symmetric Fibonacci search is the only way to achieve the maximum value
ACKNOWLEDGEMENT
The authors are grateful to the referee for his valuable suggestions. This
research was supported in part by the National Science Foundation.

REFERENCES
1. J. Kiefer, Proc, Amer, Math. Soe. , 4, 502 (1953).
2. R. E. Bellman and S. E. Dreyfus, Applied Dynamic Programming (Princeton University P r e s s , Princeton, New Jersey, 1962).
3.

L. T. Oliver and D. J. Wilde, "Summetric Sequential Minimax Search for


an Optimum," Fibonacci Quarterly, 2, 169 (1964).

4. D. J. Wilde, Optimum Seeking Methods (Prentice-Hall, Inc. , Englewood


Cliffs, N.J. (1964), pp 24 41.
* *

ALGEBRA THROUGH PROBLEM SOLVING, Book Review, Cont'd from p. 264.


From the standpoint of the Fibonacci association, this text is a landmark
in its recognition of the pedagogical value of the Fibonacci series. Let us hope
that other authors will see the wisdom of incorporating such interesting and
pregnant material into their textbooks0
it i, ir *

The Fibonacci Association invites Educational Institutions to apply for Academic


Membership in the Association, The minimum subscription fee is $25 annually.
. (Academic Members will receive two copies of each issue and will have their
names listed in the Journal.

AN EXPRESSION FOR GENERALIZED FIBONACCI NUMBERS


David Es Ferguson, Prog ram ma tics In corpora t e d ,

Los Angeles, C a l i f .

An interesting expression for the Fibonacci numbers is presented here


which relies on the modulo three value of the subscript.
a-i,./2i + a - l \
_. .
F3a = 1 - Y I
(-lf'l8l
i=o \
3i
/

(la)

(a > 0)

(lb)

a-i '/2i + a \
F 3 a + 1 = 1 - 2 V I.
(~l)a~V
M \3i + 1/

(a > 0)

(1c)

a-i / 2i + a + l \
F 3 a + 2 = 1 - 4 V ( "" ' ~
" 1 (-if" 4 8 l
i=0 V 3 1 + 2
/

(a > 0)

This is a special case of a more general expression for the generalized Fibonacci
numbers [ 1] .
(2a)

V^m = 1

(m = -n + l f . . . , 0 )

(2b)
K
'

n
V
= Y V
,
n5m
La n,m-k
fc=i

It is seen that F m = V2 m - 2 *
It is interesting to note that these numbers arise in the analysis of polyphase
merge-sorting with n + 1 tapes.

The Vn m

represent the total number of

strings on all the tapes and also the length of strings (assuming initial length
of 1) at each step of the polyphase merge process. A description of the polyphase merge-sort can be found in [ 2 ] ,
The general expression can be written ass
a / i n + a
<*>

n,a(n + l) + b =

1 +

-V-DE

+ b - l \

).(-D art (2^ 1 ) 1

a
i= o \
" *
/
(b = l , - - - , n + 1), (a > 0 ) .

270

[Oct. 1966]

AN EXPRESSION FOR
GENERALIZED FIBONACCI NUMBERS

271

Let
fnx(x)l = *-*
Y Vn m x m
?
m=o
It follows immediately that
(1 - x - x2
x n )f n (x) = 1 + (n - l)x + (n - 2)x 2 + + x11"1
Therefore
1 + (n - l)x + (n - 2)x2 + + x11""1
1, - _x - _.
x22 - - __n
x"

1 + (n -'2)x - x2 - - x n
1 - 2x + x n + 1

(4)

1 - 2x + x n + 1

1 - x
If

= y w
1 - 2x + xn+l
the sequence W
is defined by:
J
n,m
W
(5a)
n?m =
(5b)

x
n,m

(m < 0)

Wnjm = 1

(5c)
x
'
From Eq. (4)

(m = 0)

W
= 2W
- W
n,m
n,m-i
n,m-n-i
V^m
= 1 + (n - 1)W
^

(m > 0)

(m > 0)

Theorem:
<6>

na(n+0+b=
n,a(n+i)+b
j=0

2 b + (n + l ) j

( n + 1 ) j + b

^a-j

(This formula immediately yields the identity '(3).)


Proof:

1 - 2x + x

" = E (2x - x n + n m

n+1

m=o

OO

CO

E
E ( i 9 k / -jxm-k x (n+i)m-nk
m=o k = o \ k / z {~1}

[Oct.

AN EXPRESSION FOR

272

Rearranging this sum in terms of powers of x, let (n + l)a + b = (n + l)m nk. It follows that k = b (mod (n +1)), so k = (n+ l)j + b for some j > Q*
Changing the sum on m and k into a sum on a? b and j s and noting that
m = a + b + nj, results in:
OO

OO

x(n+i)a+b /

a=o b=o

+ b + n3

| 2 ( n + 1 + b (-i) a -J

j=o \ (n + 1) j + b /

This completes the proof. A similar method was used by Polya [3 ] to solve
another recurrence relation.
Another interesting expression which arises from this analysis is the general
expression for the numbers defined by:
(7a)

Un?m - 0

(7b)
v
'

U
- 1
n.m
m
U
= T u
.
n.m
-> n.m-i

(7c)
v
'

It is seen that F

(m = -n + l , - - - f - l )
(m = 0)
'

(m > 0)
'

i=l

= U? r n _ 1 .

These numbers also arise in the analysis of polyphase merge-^sorting;


they represent the number of strings produced at each step of the process.
The general expression i s :
a

(/in+a + b\

/in+a + b - l \ )

w-^K...) (..,-x )r '


(b = (), ,n) , (a + b > 0)
The proof is similar to the above and is omitted*
ACKNOWLEDGEMENT
The author is indebted to Donald E. Knuth for suggesting the proof used above.

1966]

GENERALIZED'FIBONACCI NUMBERS

273

REFERENCES
1. Ee P. Miles, J r . , "Generalized F i b o n a c c i Numbers a n d Associated
Matrices/' Amer, Math. Monthe 67 (October I960), pp 745 752.
2e Re L. Gilstad, "Polyphase Merge Sortingan Advanced Technique/' Proe.
Eastern Joint Computer Conference, Dec. I960.
"
3. G. P'olya, Induction and Analogy in Mathematics, Princeton,
p. 77.

Chapter 5

FIBONACCI YET AGAIN


J. A. He Hunter
Consider a triangle such that the square of one side equals the product of
the other two sides.
Then we have sides: X, \/XY ? and Y; say X > Y.
Eliminating an common factor we may set X = a2, Y = b 2 , so that the
"reduced" sides become a2? ab ? b 2 .
Then, for a triangle, we must have ab + b 2 > a2 which requires (\^T
l ) / 2 < b / a < (\/5 + l ) / 2 .
Hence a sufficient condition for a triangle that meets the requirements is
F2n1/F2n< b/a < F2n/F2n1

with

X = ka 2 , Y = kb2

All subscription correspondence should be addressed to Brother.U. Alfred,


St. Mary ! s College, Calif. All checks ($4.00 per year) should be made out to
the Fibonacci Association or the Fibonacci Quarterly.

Manuscripts intended

for publication in the Quarterly should be sent to Verner E. Hoggatt, Jr. ,


Mathematics Department, San Jose State College, San Jose, Calif. All manuscripts should be typed, double-spaced. Drawings shouldbe made the same size
as they will appear in the Quarterly, and should be done in India ink on either
vellum or bond paper.
the editors.

Authors should keep a copy of the manuscripts sent to

A POWER IDENTITY FOR SECOND-ORDER RECURRENT SEQUENCES


V . E . Hoggatt, J r . , San Jose State College, San Jose, C a l i f .
and D . A . Lind, University of Virginia, Charlottesville, V a .

1.

INTRODUCTION

The following hold for all integers n and k:


F

n+k ~ F k F n+i

+ F

k-lFn

n+k = k , )

( F k F k- 2 ) F n + 1 " ( W k - ^ '

n+k = ( F k F k - i F k - 2 / 2 > F n + 3

< F k F k-i F k-3) F n + 2 " ( F k F k- 2 F k- 3 ) F n + i

" < F k-i F k- 2 F k- 3 /2K

These identities suggest that there is a general expansion of the form


jjy

(1.1)

n+k E a J

iKP)F

3=o

Here we show such an expansion does indeed exist, find an expression for the
coefficients a.(k s p), and generalize (1.1) to second order recurrent sequences.
2S

A FIBONACCI POWER IDENTITY

Define the Fibonomial coefficients


F F
F
m m-i
m-r+i
F1F2--Fr

ml
:\ by

(r > 0);

Jarden [4] proved that the term-by-term product z


obeying the Fibonacci recurrence satisfies
274

= 1

of p sequences each

[Oct. 1966]

A POWER IDENTITY FOR


SECOND-ORDER RECURRENT SEQUENCES
p+i

p+1

J2 H-> 3 ( J + I ) / 2

(2.D
for i n t e g r a l n.

275

n-j

In p a r t i c u l a r z n = F ^ obeys (2.1). C a r l i t z , [ 1 , Section 1] has

shown that the d e t e r m i n a n t


D

"n+r+s

(r,s =

0,l,-,p)

has the value


Dp

= (-l)P(P+1)(n+1)/2|(P).

( F ^ . - F p ) ^

i=o \ J /
implying that the

p+1

sequences

independent over the r e a l s .

{F },{F P

} , , {FP

are

linearly

Since each of t h e s e s e q u e n c e s obeys the (p + 1)

o r d e r r e c u r r e n c e r e l a t i o n (2.1), they m u s t s p a n the space of solutions of (2.1).


T h e r e f o r e an expansion of the f o r m (1.1) e x i s t s .
To evaluate the coefficients
giving a.(k,p) = 6.7
fined by

<5., = 0

for
if

a.(k,p) in (1.1) we f i r s t put

0 < j?k < p,

j 4= k,

6-,,

k 0,ls , p ,

w h e r e 6.-. i s the K r o n e c k e r - d e l t a d e -

= 1.

Next we show that the s e q u e n c e

{a.(k,p)}l^o
obeys (2.1) for j = 0 , 1 , , p .

Indeed from (1.1) we find

p+i

o = J2 <-D r(r+1)/2
p I p+i
j=o j r=o

P + 1
"n+k-r

P+1
r

a ( k - r , p ) } F;
n+j

But the F v . (j = 0 , 1 , e * e ,p) a r e l i n e a r l y independent, so that


n+j

1 r(r+1)/2

E <- >
r=o

p+1
r

a.(k - r , p ) = 0 ( j

0,l,---,p)

276

A POWER IDENTITY FOR

Now consider

b.(k,p) = (F, F,

F,

)/F,

[Oct.

.(F.F. " F F

>00

F.

for

j = 0, ! , ,p - 1, b (k,p) = I p j , together with the convention that F 0 / F 0 = 1.


Clearly bj(k,p) = <5., for 0 < j,k < p.

Since {b.(k,p)}j_

is ;the term-by -

term product of p Fibonacci sequences, it must obey (2.1). Thus {a.(k,p)},


oo
th
^
and |b.(k,p)},
obey the same (p + 1) order recurrence relation and have
their first p + 1 values equal (j = 0 , l , - - , , p ) , so that a.(k,p) = b.(k,p).
J
r
Since F_ = (-l) n+1 F , it follows that
F -F.
= F .F1V
1 (-l)
-1
j-p
p-J
'

(P~J)(P-J+3)/2

so that for j = 0 , 1 , ,p - 1, we have

+3
a.(k,p) = (-1) (lH)(P-J )/2

(-i)

I VW" F 1 ll(Fr-Fi)(V]"-^flFk-j/

(P-J)(P-J+*)/2

which is also valid for

ClDf]

(F

k-P/Fk-j>'

j = p using the convention

F0 / F 0 = 1.

Then (1.1)

becomes
P
(2 2)

'

n+k = 2 ^ "

v(p-3)(p-J+3)/21

ir

LP.

for all k We remark that since consecutive p

<Fk^/Fk-j>^

powers of the natural num-

bers obey

5 ( "" i r J ( P - 1 ) ( n + J ) P = '
3=o

a development similar to the above leads to

(2.3)

(n + k)*-

e l )

2^
<-'(:)(;)(-)
j=o

(n + if

1966]

SECOND-ORDER RECURRENT SEQUENCES

277

a result parallel to (2.2)


3.

EXTENSION TO SECOND-ORDER RECURRENT SEQUENCES

We now generalize the result of Section 2.

Consider the second-order

linear recurrence relation


(3.1)

yn+2 =

Pyn+1

-qyn

(q * 0)

Let a and b be the roots of the auxiliary polynomial x2 - px + q of (3.1),


Let w

be any sequence satisfying (3,1)

(a - b) if a + b, and u

= na

and define u

by u

= (an-bn)/

if a = b s so that u

also satisfies (3.1).


fml
Following [ 4 ] , we define the u-generalized binomial coefficients I I
by
T
u u
* u
,,
i
r
n

m l = _m_m-i
rn^r+i
Tml =
rju
UiU 2 ---u r
|_0ju
Jarden [4] has shown that the product x of p sequences each obeying (3.1)
satisfies the (p + 1)

order recurrence relation


p+i

(3.2)

^TVl)^" 1 )/ 2

p+1
J

J=o

If all of these sequences are w ,

^-3=

then it follows that x

= w p obeys (3.2).

It is our aim to give the corresponding generalization of (1.1) for the


sequence
w ; that i s , to show there exists coefficients a.(k,p,u) = a.(k) such
le
that

(3 3)

'

^ + k = I ] a 3 ( k ) w Jn+j
j=o

and to give an explicit form for the a.(k)e

V W ) = K+r+sl

Carlitz [ 1 , Section 3 ] proved that

<r's = O.L.P)

278

A POWER IDENTITY FOR

[Oct.

is nonzero, showing that the p + 1 sequences

are linearly independent

Reasoning as before, we see these sequences span

the space of solutions of (3.2), so that the expansion (3.3) indeed exists.
ting k = 0 , l , - * * , p in (3.3) gives a.(k) = 6.,

for 0 < j , k < p.

Put-

It also fol-

lows as before that the sequence

satisfies (3.2). Now consider


b.(k.p.u) = b.(k) = u, u, / u , /u,- .(u.u. / u 1 u_1 1 *-*u. )
J
J
k k-i
k-p' k-f j j-i
I
j~p'

for j = 0 , l , . . - , p - l , bp(k) = [^ , along with the convention u0 /u 0 = 1


Then bj(k) = 6.-. for 0 < j , k < p. Also{b-.(k)}, obeys (3.2) because it
is the product of p sequences each of which obeys (3.1). Since {a.(k)V?
th
3
^~
and {b-(k)}?
(j = 0,l,,p) obey the same (p + 1)
order recurrence
relation and agree in the first p + 1 values, we have a.(k) = b.(k). Now
ab = a, so that u^ = (a" n - b ~ n ) / ( a - b) = - q n u . Then

~l

j-p

p-j

and thus for j = 0 , 1 , ,p - 1 we see


(3.4)
a (k) =

( .i)P-J q (P-J)(P-J

+1

/ U, U,

)/2 /
\

U,

k k-i
Vp-l'

= (-l)P"jq(p"j)(p~3+l)/2
.PJuu.

,J V /

U U

UJ

\ / U,

k-p+i W
pp-i
* . ]/
" U l ' V ( u j- * * u i>< u p-j-'' ul> /1
(u,
/u, .)
v
k-p k - j '

k-p\
u
k-j /

,
'

which is also valid for j = p using the convention u0 /u 0 = 1. Therefore (3.3)


becomes
p
35

<->

.-ki

( 1)P

n+k=Z -

1
J ,<p-i)(p-J+ )/*" '
LPJ

(u,
/u, .) w ,.
v
k - p / k - j ' n+j
u

1966

SECOND-ORDER RECURRENT SEQUENCES

279

Carlitz has communicated and proved a further extension of this result.


Let
(P)
n
where the a.

(3.6)

n+aj n+a2

n+a r

are arbitrary but fixed nonnega,tive integers.

-r n + k ^ E ^

P-Jq(P-J)(P-J + 1)/2

Then we have

1(u.

;
k-p /u.k - j.)x
n+j

9,

3=o

where u0 /u 0 still applies. We note that putting aA = a2 = e = a

=0

re-

duces (3.6) to (3.5).


To prove (3.6) using previous techniques requires us to show that the
sequences

{?'}{;}..{"&}
are linearly independent.

To avoid this, we establish (3.6) by induction on k.

Now (3.6) is true for k = 0 and all n. Assume it is true for some k ^ 0 and
all ns and replace n by n + 1 ,

5>

x(p)
xi+k+i

u j=o
p
J

giving

1 > : P-j (P-J)(P-J+0/2

k-p X(P)

1)P P-J+UP-J+I)(P-J
j+lq

X> ~

+2

n+j+1

Vj
)/ 2

1-1

UJ=1

X(P) +
J

k-i+i
J

It follows from (3.2) that


p+i
K(P)

n+p+i

( 1)jq3(j 1)/2

"S "

"

p + 1

P-L(P-J)(P-J )/
J^o

n+p+i-j

j
+1

X(P)

u
p+1
3

n+j
-u

,( P )
n+p+i

280

[Oct

A POWER IDENTITY FOR

Thus

1 k

(P)
x
n+k-H

1)P-jq(P-j)(P"j+l)/

JP)

3 k-j+i

3 -1

-*U j=0

p+i k-j+i

k-p 3

Since

qP 3+lu

ViVj+i -

"

k-p u j

ViVj+i

'

we have

Z ( - 1)P " 3q

JP)

P-L(P-j)(P-J+i)/2

n+k+i - u,k+il

P
( 1) P~J q (P~J)(P-J+

u.u, . , ,
3 k-j+i

J - 1

u j=o

k + 1i

JVitL x (p)

)/ 2 P

u.

u.k-p+i
Vj+i

n+jJ

(p)
n+j

J . i=0

completing the induction step and the proof e


4.

SPECIAL CASES

In this section we r e d u c e (3.5) to a g e n e r a l Fibonacci p o w e r Identity and


to an identity involving p o w e r s of t e r m s of an a r i t h m e t i c p r o g r e s s i o n .
if we l e t
s 4 0,

, , u = F , where r
n
ns+r'
n
ns'
then both w and u
satisfy

(4.1)

= F

vr

n+2'

- L yJ
+ (-1) v
s n+i

= 0 .

The r o o t s of the a u x i l i a r y polynomial of (4.1) a r e distinct for


w

and u

satisfy the conditions of the p r e v i o u s section,,

g e n e r a l i z e d binomial coefficients
Fml
coefficients
,
defined by

s ==
f 09

Ju

s o that

In this c a s e the u -

b e c o m e the s - g e n e r a l i z e d
L

First

and s a r e fixed i n t efog e r s with

Fibonacci

1966]

281

SECOND-ORDER RECURRENT SEQUENCES


F

F
1
ins (m-i)s
(m-t+i)s (t > 0) ;
e
F F
F
ts t s - s
s

= 1

A recurrence relation for these coefficients is given in [ 3 ] , Now here


q = <-l) S , SO ( . D P - J q ^ M P - ^ / 2

ei)(PJ)[s(P^lH]/2

Then (3.5) yields


p

ral)(P-J)[s(p-j-M)+2]/2

(4.3) F(n+k)s+r

F
F

.j J.

1=0

(k-p)s

(k-J)s

Fp

<n+s+r

Putting s = 1 and r = 0 gives equation (2,2).


On the other hand, if we let w = ns + r and u - n, where r and s
'
n
n
'
are fixed integers, then w and u obey
JV

J
n+2 - 2v
n-H + JVn

(4.4)

=0

Since the characteristic polynomial of (4.4) has the double root x = 1, both
w

and u

satisfy the conditions for the validity of (3.5). In this case we have

q = 1 and J , j = | , 1, the usual binomial coefficient.

(4.5) ([n + k]s + r)

pf

\p/\j/\k - j .

Then (3.5) becomes

(O + j]s + rf

This reduces to (2.3) by setting s = 1 and r = 0S

REFERENCES
1. L9 Carlitz, "Some Determinants Containing Powers of Fibonacci Numbers s?f
Fibonacci Quarterly, 4(1966), No, 2, pp 12 9 - 1 3 4 .
20 V. E0 Hoggatt, Jr a and A P 0 Hillman, "The Characteristic Polynomial
of the Generalized Shift Matrix/ 1 Fibonacci Quarterly, 3(1965), No 25
pp 91 94.

282

A POWER IDENTITY FOR


SECOND-ORDER RECURRENT SEQUENCES

[Oct. 1966]

3. Problem H-72, Proposed by V. E. Hoggatt, J r . Fibonacci Quarterly,


3(1965), No. 4, pp 2 9 9 - 3 0 0 ,
4. D. Jarden, "Recurring Sequences, n Riveon Lematimatika, Jerusalem
(Israel), 1958, pp 4 2 - 4 5 .
5. Roseanna F. Torretto and J. Allen Fuchs, ? Generalized Binomial Coefficients ," F i b o n a c c i
2(1964), Noe 4, pp 2 9 6 - 3 0 2 .
ACKNOWLEDGEMENT
The second-named author was supported in part by the Undergraduate
Research Participation Program at the University of Santa

Clara

through

NSF Grant GY-273,


* *

REQUEST
The Fibonacci Bibliographical Research Center desires that any reader
finding a Fibonacci reference send a card giving the reference and a brief
description of the contents. Please forward all such information to;
Fibonacci Bibliographical Research Center,
Mathematics Department,
San Jose State College,
San Jose, California.
* * *

The Fibonacci Association invites Educational Institutions to apply for Academic


Membership in the Association. The minimum subscription fee is $25 annually.
(Academic Members will receive two copies of each issue and will have their
names listed in the Journal.
*****

ELEMENTARY PiOBLEiS AND SOLUTIONS


Edited by A . P . HiNman/University of New Mexico, Albuquerque, New Mex*

Send all communications regarding Elementary Problems and Solutions to


Professor A. P. Hillman, Department of Mathematics and Statistics, University of New Mexico, Albuquerque, New Mexico 87106* Each problem or solution should be submitted in legible form, preferably typed in double spacing,
on a separate sheet or sheets, in the format used below.

Solutions should be

received within three months of the publication date,


B-94

Proposed by Clyde A. Bridge^

Show that the number N

Springfield

Jr. College, Springfield,

Hi

of non-zero terms in the expansion of

....

b2

...

a3

b3

...

-1

an

is obtained by replacing each a. and each b^ by 1 and evaluating K^

Show

further that N n = F n + 1 , the (n + l)st Fibonacci number,


B-95

Proposed by Brother tf. Alfred, St. Mary*$ Co//-ege,.Ccr/if.

What is the highest power of 2 that exactly divides


F

B-96

1F2F3

Proposed by Phil Manaf University

100

of New Mexieo% AlbuquerquetNew

Mex*

Let G n be the number of ways of expressing the positive integer n as


an ordered sum a t + a2 + * + a s with each aj in the set { l , 2 , 3 } 0 (For
283

[Oct.

ELEMENTARY PROBLEMS

284

example, G3 = 4 since 3 has just the expressions 3,2 + 1, 1 + 2, 1 + 1 + 1.)


Find and prove the lowest order linear homogeneous recursion relation s a t i s fied by the G n .
B-97

Proposed by Douglas Lind, University of Virginia, Charlottesville,

Let A = {a^},

Vb.

be an increasing sequence of numbers and let A(n) de-

note the number of terms of A not greater than n. The Schnirelmann density
of A is defined as the greatest lower bound of the ratios

A(n)/n for n = 1,

2,- . Show that the Fibonacci sequence has density zero.


B-98

Proposed by Douglas Lind, University o f Virginia, Charlottesville,

Let F

be the n

Vb.

Fibonacci number and find a compact expression for

the sum Sn(x) = F t x + F2x2 + F3x3 + + F n x n Q


B-99

Proposed

by Douglas Lind, University of Virginia, Charlottesville,

Va.

Find a compact expression for the infinite sum


S2(x)

T(x) = st(x) + ~^r-

S3(x)

-jr~

*9 >

where S (x) is as defined in B-98.

SOLUTIONS
DIFFERENCE AND DIFFERENTIAL EQUATIONS
B-76

Proposed by James A. Jeske,San Jose State Colleget

San Jose, Calif.

The recurrence relation for the sequence of Lucas numbers is L n + 2 Ln+i - Ln = 0 with Lj = 1, L2 = 3.
Find the transformed equation, the exponential generating function, and
the general solution.

1966]

AND. SOLUTIONS

Solution by Douglas Unci! University

of irginias

285
Vb.

Charlottesviiieg

By (4.2) of Jeske f s a r t i c l e , the transformed equation is

L2(D)Y = Y" - Yf - Y = 0, Y(0) = 2, Yf(0) = 1

Now rt = (1 + /s/5)/2 and r 2 = (1 - V5)/2 are the roots of L 2 (r) = 0, hence


with the given initial values we may determine the solution for the Lucas s e quence to be

i, - (H*r (H*)T .
The exponential generating function for the Lucas sequence is thus
OO

Yft) - e * + a* = 2 X

...

n=o
Also solved by Clyde A. Bridget, Howard

I.Walton,

and the

Proposer.

IT PAYS TO CHECK
B-77

Proposed by James A. Jeske,San Jose State Colleget

San Joeg

Calif.

Find the general solution and the exponential generating function for the
recurrence relation
yn+3 - 5yn+2

8yn+i -

yn

with y0 = 0 , yt = 0, and y 2 =-1.


Solution by Douglas Unds University

Since

of Virginiat

Charlottesville;

Vb.

L3(x) = x3 .- 5x2 + 8x - 4 = (x - l)(x - 2)(x - 2) = 0, we have,

using the notation of section 4 of Jeske f s paper,

r 1 = 1, nij = 1, r 2 = 29

286

ELEMENTARY PROBLEMS

m 2 = 2,

and m = 2e

[Oct.

With the given initial v a l u e s we find the exponential

g e n e r a t i n g function to be

Y(t) = - e t + e 2 t ( l - t)
which, by applying the i n v e r s e t r a n s f o r m (3.3), y i e l d s the g e n e r a l solution a s

Also solved by Clyde A, Bridget and the Proposer. Douglas Lind also noted that formula (4.8) f Jeske's paper is not correct.
A LUCAS SUM
B-78

Proposed by Douglas lind, University of Virginia, Charlottesvillef Va.


Show that
F

= L
n

+L
n-2

++ L
n-6

+ e
n-2-4m

w h e r e m i s the g r e a t e s t i n t e g e r in (n - 3 ) / 4 ,
e

,
n '

and e

n > 2 ,
*

= 0 if n = 0 (mod 4),

= 1 if n t 0 (mod 4).

Solution

by the

Proposer.

Proof by induction: The p r o p o s i t i o n i s e a s i l y shown t r u e for n = 3 , 4 , 5 , 6 .


Now a s s u m e the t h e o r e m t r u e for

Fk-f4 =

3 ^ n ^ k + 3.

Then one finds that

k+2 + F k = Lk+2 + (Lk-2 + Lk-e + * 8 * +

s o that the t h e o r e m i s t r u e for


proof 0
Also solved by David

Zeitiin,

k + 4,

k~2-4m

k)

completing the induction s t e p and the

1966]

AND SOLUTIONS

287

AN ALMOST LUCAS SUM


B-79

Proposed by Brother U. Alfred, St Mary's


Let a = (1 + V 5 ) / 2 .

Colleges Calif.

D e t e r m i n e a c l o s e d e x p r e s s i o n for

= [<!

[a2]+..-+[an] ,

w h e r e the s q u a r e b r a c k e t s m e a n " g r e a t e s t i n t e g e r in?T


Solution

by the

Proposer.

If b = (1 - \ / 5 ) / 2 ,
[ a k ] = L,

= L, - b

with b negative and | b | < l a

if k- i s odd and [ a ^ ] = L, - 1 if k i s even,

Hence

It follows that

n
X

i C Lk ~ \J

Va " 3

k=i
Also

solved

by J.L. Brown/ Jr. and Jeremy C. Pond.


OUR MAN OF PISA

B-80 Proposed by Maxey Brooke, Sweeny, Texas.


Solve the division alphametic

FIB

_PISA
XONACCI

w h e r e each l e t t e r r e p r e s e n t s one of the nine digits 1,2, , 9 and two l e t t e r s


may r e p r e s e n t the s a m e digit.
Solution

by the

Proposer.
9854
382 [ 3 7 6 4 2 2 8

T h i s solution may not be unique.

288

ELEMENTARY PROBLEMS AND SOLUTIONS

[Oct

1966]

GAUSSIAN PRIMES
B-81

Proposed b y Douglas LiJid, University of Virginiag Charlottesville/


P r o v e that only one of the Fibonacci n u m b e r s l f 2 s 3 ? 5 ? o s e

Va.

i s a p r i m e in

the r i n g of G a u s s i a n i n t e g e r s .

Solution by L. Carl/tz, Duke University, Durham, NX.


Since
F 2 n+i = F n + 1

+ Fn

= (Fn+1 + Fni)(Fn+1 - F Q i )

it follows that F 2 n + i i s c o m p o s i t e in the G a u s s i a n r i n g for all n > 0


F2n = F L

it follows that

F2n

Since

i s c o m p o s i t e in the r i n g of i n t e g e r s (kndv

t h e r e f o r e in the G a u s s i a n ring) for n > 2a

F o r n = 2 S F 4 = 3; for n = 1,

F 2 = 1.
Also solved by Sidney Krqvitz and the proposer.

NOTICE TO ALL SUBSCRIBERS!!!


P l e a s e notify the Managing Editor AT ONCE of any a d d r e s s change*

The P o s t

Office D e p a r t m e n t , r a t h e r than forwarding m a g a z i n e s m a i l e d t h i r d c l a s s , s e n d s


t h e m d i r e c t l y to the d e a d - l e t t e r office.

Unless the a d d r e s s e e specifically

re-

q u e s t s the Fibonacci Q u a r t e r l y -to be forwarded at f i r s t c l a s s r a t e s to the new


a d d r e s s , he will not r e c e i v e it. (This will usually c o s t about 30 c e n t s for f i r s t c l a s s p o s t a g e . ) If p o s s i b l e , p l e a s e notify us AT LEAST THREE WEEKS PRIOR
to publication d a t e s : F e b r u a r y 15, A p r i l 15, October 15, and D e c e m b e r 15 e